軍研究國際智慧無人系統技術應用及發展趨勢
現代英語：
With the accelerated application of cutting-edge technology in the military field, intelligent unmanned systems have become an important part of modern warfare. The world’s major military powers attach great importance to the application of intelligent unmanned system technology in the military field. In the future, intelligent unmanned systems will have a profound impact on combat methods and subvert the rules of war. As a culmination of cutting-edge science and technology (such as artificial intelligence, intelligent robots, intelligent perception, intelligent computing, etc.), intelligent unmanned systems represent the highest level of development of a country’s scientific and technological strength. Therefore, research in the field of intelligent unmanned systems can greatly promote the development of existing military and livelihood fields.
At present, unmanned system equipment has emerged in military conflicts. For example, in the conflict between Turkey and Syria, Turkey used the Anka-S long-flight drone and the Barakta TB-2 reconnaissance and strike drone equipped by the Air Force to attack the Syrian government forces; the Russian Ministry of Defense also announced that militants in Syria used drones carrying explosives to launch a cluster attack on its military bases; in 2020, the United States used an MQ-9 “Reaper” drone to attack a senior Iranian military commander and killed him on the spot. Unmanned combat is coming, and intelligent unmanned systems, as a key weapon on the future battlefield, will determine the victory of the entire war.
Image from the Internet
The development of intelligent unmanned systems will not only promote the upgrading and progress of existing military technology, but also drive the intelligent development of civilian technology, including intelligent transportation systems, smart home systems, intelligent manufacturing systems and intelligent medical systems. In order to develop intelligent unmanned systems more scientifically and rapidly, major scientific and technological powers have introduced a series of plans and routes for the development of intelligent unmanned systems, striving to seize the initiative and commanding heights in the development of intelligent unmanned systems. Related ones include the United States’ integrated roadmap for autonomous unmanned systems, Russia’s national weapons and equipment plan, the United Kingdom’s defense innovation technology framework, China’s new generation of artificial intelligence development plan, and Japan’s medium- and long-term technology plan.
In recent years, from air to space, from land to sea, various types of intelligent unmanned systems have emerged in large numbers. The world’s major powers have gradually deployed intelligent unmanned systems into the military, and in some regional conflicts and anti-terrorism battlefields, the key role of intelligent unmanned systems is increasing. Therefore, this article will focus on the military needs of the future battlefield, based on the challenges of the actual complex environment faced by the future battlefield, analyze the key technologies required for the development and application of intelligent unmanned systems, and analyze the key technologies of individual enhancement and cluster enhancement from a military perspective, and explain the development trend of intelligent unmanned systems.
- Current research status at home and abroad
The concept of intelligent unmanned system has only been proposed recently. At present, its research is still in its early stages, and there is no unified definition in the world. It is temporarily defined as: an organic whole composed of an unmanned platform and several auxiliary parts, with the ability to perceive, interact and learn, and capable of autonomous reasoning and decision-making based on knowledge to achieve the goal. Intelligent unmanned systems can be divided into three major parts: land unmanned systems, air unmanned systems and marine unmanned systems according to the spatial scope of their functions. Among them, land unmanned systems mainly include reconnaissance unmanned vehicles, transport unmanned vehicles, combat unmanned vehicles, obstacle removal unmanned vehicles, bomb disposal unmanned vehicles, unmanned vehicle formations and command systems, etc.; air unmanned systems mainly include reconnaissance drones, combat drones, logistics transport drones and drone formations, etc.; marine unmanned systems mainly include reconnaissance unmanned boats, combat unmanned boats, logistics transport unmanned boats, patrol search and rescue unmanned boats, reconnaissance unmanned submarines, combat unmanned submarines and shore-based support systems, etc. This section will explain the current research status of intelligent unmanned systems at home and abroad from the above three parts.
⒈ Current status of foreign intelligent unmanned system research
⑴ Land unmanned system
Land unmanned systems are mainly used in intelligence collection, reconnaissance and patrol, mine clearance and obstacle removal, firepower strike, battlefield rescue, logistics transportation, communication relay and electronic interference. As the advantages of land unmanned systems in combat become more and more prominent, research on them has attracted more and more attention from various countries.
The United States launched the “Joint Tactical Unmanned Vehicle” project in November 1993, which is the predecessor of the “Gladiator” unmanned combat platform project. In 2006, the United States completed the design of the entire system of the “Gladiator” unmanned combat platform and officially equipped the Marine Corps in 2007. The “Gladiator” tactical unmanned combat platform is the world’s first multi-purpose combat unmanned platform. It is equipped with sensor systems such as day/night cameras, GPS positioning systems, and acoustic and laser search systems. It is also equipped with machine guns, submachine guns, tear gas, sniper systems, biological and chemical weapons detection systems, etc. It can perform reconnaissance, nuclear and biological weapons detection, obstacle breakthrough, anti-sniper, firepower strike and direct shooting in different weather and terrain.
The Gladiator unmanned combat platform is equipped with a highly mobile and survivable chassis. For this platform, a portable handheld control system has also been developed, and a series of development work has been completed around the technical issues of the control system’s anti-interference, network interoperability, miniaturization and ease of operation. However, due to the weak armor protection capability of the Gladiator unmanned combat platform and the poor concealment of its mission, its long-range reconnaissance and control system faces more interference. In addition, the US Army has also put some other land unmanned systems into service, such as the Scorpion robot and the Claw robot. In 2017, the US Army formulated the Robotics and Autonomous Systems (RAS) Strategy, which provides a top-level plan for the construction of unmanned combat capabilities. Figure 1 shows the US land unmanned system.
Figure 1 US land unmanned system
Israel, Russia, the United Kingdom and Germany have also successively carried out the development of land unmanned systems and developed a series of advanced products. The product list is shown in Table 1. For example, the “Guardian” series of autonomous unmanned vehicles developed by Israel can combine the sensors and fusion algorithms on board to autonomously detect and identify dangerous obstacles, and perform patrol, surveillance and small-scale fire strike tasks; the MARSA-800 unmanned vehicle developed by Russia can perform tasks such as transportation and logistics support, tracking and surveillance, and can realize autonomous path planning and avoid obstacles during the execution of tasks. The unmanned vehicle has been deployed on the Syrian battlefield. The United Kingdom and Germany also started research on land unmanned systems earlier. The United Kingdom launched a trolley bomb disposal robot in the 1960s, and later launched the Harris T7 tactile feedback robot for performing dangerous tasks such as bomb disposal and bomb disposal; the “Mission Master” ground armed reconnaissance unmanned vehicle developed by Germany’s Rheinmetall is mainly used to perform tactical surveillance, dangerous object detection, medical evacuation, communication relay and fire support tasks.
Table 1 Land unmanned systems of various countries
⑵ Aerial unmanned systems
Aerial unmanned systems are mainly based on single drone platforms and drone clusters. Due to their advantages such as wide field of view, freedom of flight, and good equipment carrying capacity, drones are widely used in the military field and have played a great role in military conflicts in recent years. The main functions of aerial unmanned systems include: intelligence gathering, reconnaissance and surveillance, decoy target aircraft, target tracking, tactical strikes and air rescue.
In 2000, the U.S. Air Force Research Laboratory proposed the concept of autonomous combat for unmanned aerial vehicles, quantified the degree of autonomy of unmanned aerial vehicles, and formulated a development plan. The quantitative content and development stage of the degree of autonomy of unmanned aerial vehicles are shown in Figure 2.
Figure 2 Autonomous control level and the trend of autonomous
unmanned aerial vehicles In 2003, the United States merged the unmanned combat aircraft system projects of the Air Force and the Navy, launched the “Joint Unmanned Combat System” (J-UCAS) project, and began research on the unmanned combat aircraft X-47B. In 2006, the U.S. Navy proposed the “Navy Unmanned Combat Air System” (N-UCAS) project, which aims to introduce unmanned combat aircraft to the aircraft carrier-based aircraft wing and continue to conduct research on the X-47B. Between 2012 and 2014, the aircraft carrier catapult, landing, touch-and-go and other tests were completed many times, and the autonomous aerial refueling test was completed in 2015. The X-47B attack drone is an autonomously maneuverable, stealthy, and land-based and ship-based unmanned combat aircraft. It has the characteristics of high range and high flight time, and is equipped with advanced sensors such as illumination radar, optoelectronic guidance system, and aperture radar. Its main functions include intelligence reconnaissance, target tracking, electronic warfare interference, and firepower strikes. Other unmanned aerial systems developed by the United States, such as the Global Hawk, Predator, Hunter, and Raven, have also been in service in the military, as shown in Figure 3.
The “Harpy” drone developed by Israel is equipped with anti-radar sensors, optoelectronic guidance systems and missiles, and can autonomously attack enemy radar systems, as shown in Figure 3.
Figure 3 Aerial Unmanned Systems of Various Countries
A single aerial unmanned system is easily interfered with and attacked when performing a mission, resulting in mission failure, while an aerial unmanned system cluster can make up for this defect and give full play to the advantages of aerial unmanned systems. The Defense Advanced Research Projects Agency (DARPA) of the United States has successively launched the “Gremlins” low-cost drone project, the low-cost drone cluster project, the “Perdix” micro-drone airborne high-speed launch demonstration project, and the offensive swarm enabling tactics (OFFSET) project for aerial unmanned system clusters. By developing and testing the architecture, communication system and distributed control algorithm for unmanned system clusters, an autonomous control system for drone clusters has been developed, and cutting-edge scientific and technological technologies such as artificial intelligence, situational awareness, virtual reality and augmented reality have been used to enhance the comprehensive combat capability of aerial unmanned system clusters on the battlefield.
⑶ Marine unmanned systems
Marine unmanned systems include two types: surface unmanned systems and underwater unmanned systems. Among them, surface unmanned systems mainly refer to surface unmanned boats (hereinafter referred to as “unmanned boats”), which are mainly used to perform tasks such as maritime search and rescue, reconnaissance and surveillance, firepower strikes, patrol security, electronic interference, logistics support and decoy target ships; underwater unmanned systems mainly refer to unmanned submersibles. Compared with manned submarines, they have the advantages of no casualties, high concealment and high autonomy, and are mainly used to perform intelligence collection, target monitoring, combat deterrence and firepower strikes. In 2018, the US Navy released the “Navy Department Unmanned System Strategic Roadmap”, and in 2019, it released the “Navy Artificial Intelligence Framework”, which provides route planning and guidance for the development of naval operations and marine unmanned systems.
In terms of surface unmanned systems, the United States proposed the “American Advanced Concept Technology Demonstration Project” (ACTD), one of whose important tasks is to carry out research on the “Spartan Scout” unmanned boat. The project was completed in 2007 and tested in the Iraqi theater. The “Spartan Scout” unmanned boat is equipped with an unmanned driving system and a line-of-sight/beyond-line-of-sight communication system, as well as advanced sensors such as electro-optical/infrared search turrets, high-definition cameras, navigation radars, surface search radars, and global positioning system receivers, as well as weapons such as naval guns, anti-ship missiles, and anti-submarine sensors. It is mainly used to perform intelligence collection, target monitoring, information reconnaissance, anti-mine and maritime security tasks, and has a certain degree of autonomy. The “Sea Hunter” unmanned boat developed by the United States is equipped with sonar and optoelectronic sensors, as well as short-range and long-range radar detection systems and expandable modular sonar systems. It is mainly used to perform tasks such as identifying and monitoring suspicious targets and guiding fire strikes. The US marine unmanned system is shown in Figure 4. The “Protector” unmanned boat developed by Israel is mainly used to perform intelligence reconnaissance, suspicious target identification, tactical interception, electronic interference and precision strikes (Figure 4). The unmanned surface reconnaissance boat developed by Russia can perform rapid patrol tasks under the command of the mother ship and inspect and monitor designated areas to search for intelligence.
Figure 4 Marine unmanned systems of various countries
In terms of underwater unmanned systems, the nuclear-powered unmanned submarine “Poseidon” developed by Russia can carry conventional and nuclear warheads to perform reconnaissance and strategic nuclear strike missions, as shown in Figure 4. The “Knifefish” unmanned submarine developed by the United States can scan suspicious objects and search for intelligence by emitting low-frequency electromagnetic waves; the “Tuna”-9 unmanned submarine developed by the United States can carry a variety of standard payloads and can be used to perform offshore exploration, anti-mine, surveillance and reconnaissance (ISR) and other tasks.
⒉ Current status of domestic intelligent unmanned system research
In recent years, China’s military intelligent unmanned systems have developed rapidly. This article will explain the three aspects of land unmanned systems, air unmanned systems and marine unmanned systems.
In terms of land unmanned systems, the National University of Defense Technology and Sany Heavy Industry Co., Ltd. jointly developed the “Desert Wolf” land unmanned light platform, which is powered by tracks and equipped with weapon systems such as grenade launchers and machine guns. It can be used to perform logistics transportation, wounded transportation, reconnaissance monitoring, firepower strikes and other tasks. The “Longma” series of unmanned vehicles developed by Sunward Intelligent Group have strong transportation and obstacle crossing capabilities. The “Shenxing-III” military ground intelligent robot system developed by Nanjing University of Science and Technology has strong autonomous navigation and intelligence reconnaissance capabilities. The unmanned nuclear reconnaissance vehicle jointly developed by the National University of Defense Technology and Harbin Institute of Technology has high mobility and armor protection capabilities. The weapon system it carries can perform fire strikes and has certain autonomous capabilities.
In terms of aerial unmanned systems, the “Wing Loong” series of unmanned aerial vehicles developed by Chengdu Aircraft Industry Group has fully autonomous horizontal take-off and landing capabilities, cruise flight capabilities, air-to-ground coordination capabilities, and ground relay control capabilities. It is equipped with multiple types of optoelectronic/electronic reconnaissance equipment and small air-to-ground precision strike weapons, and can perform intelligence reconnaissance, target tracking, fire strikes and other tasks. The “Rainbow” series of unmanned aerial vehicles developed by China have medium-altitude and long-range navigation capabilities, can carry electronic jamming systems and a variety of weapon systems, and can perform fire strikes, intelligence reconnaissance, communication jamming, radio wave jamming and other tasks; the attack 11 type unmanned aerial vehicle developed has extremely strong stealth capabilities and can carry precision-guided missiles for ground attack missions. China’s aerial unmanned systems are shown in Figure 5.
Figure 5 China’s aerial unmanned systems
In terms of surface unmanned systems of marine unmanned systems, the “Tianxing No. 1” unmanned boat, developed by Harbin Engineering University, uses oil-electric hybrid power, with a maximum speed of more than 92.6km/h and a maximum range of 1,000km. It is currently the fastest unmanned boat in the world. The boat integrates technologies such as autonomous perception, intelligent control, and autonomous decision-making, and can achieve rapid situation information recognition and danger avoidance of the surrounding complex environment. It can be used to perform tasks such as meteorological information monitoring, landform mapping, alert patrol, intelligence reconnaissance, and firepower attack. The “Jinghai” series of unmanned boats developed by Shanghai University have semi-autonomous and fully autonomous operation capabilities, and can perform tasks such as target reconnaissance, ocean mapping, and water quality testing. The “Haiteng 01” intelligent high-speed unmanned boat developed by Shanghai Maritime University is equipped with sensors such as millimeter-wave radar, laser radar, and forward-looking sonar. It can perform suspicious target monitoring, underwater measurement, maritime search and rescue, and other tasks, and has fully autonomous and semi-autonomous navigation capabilities. The JARI intelligent unmanned combat boat developed by Jiangsu Automation Research Institute is equipped with detection equipment such as photoelectric detectors and four-sided phased arrays. At the same time, it is also equipped with weapon systems such as missiles and torpedoes, which can perform tasks such as intelligence collection, enemy reconnaissance, and precision firepower strikes. The “Lookout II” unmanned missile boat jointly developed by Zhuhai Yunzhou Intelligent Technology Co., Ltd. and other units is equipped with a fully autonomous unmanned driving system and missiles and other weapons, which can perform tasks such as enemy reconnaissance, intelligence collection, and precision firepower strikes. China’s marine unmanned system is shown in Figure 6.
Figure 6 China’s marine unmanned system
In terms of underwater unmanned systems of marine unmanned systems, the “Devil Fish” unmanned submersible developed by Northwestern Polytechnical University is a bionic manta ray unmanned submersible that has completed a deep-sea test of 1025m. The “Wukong” full-sea depth unmanned submersible developed by Harbin Engineering University has successfully completed a deep dive and autonomous operation test of 10,896m. Deep-sea submersibles such as “Qianlong No. 1” and “Seahorse” developed by China have successfully completed deep-sea exploration missions.
⒊ Summary of the current state of technology
At present, intelligent unmanned systems have been gradually applied to various fields of military applications, and with the development of cutting-edge science and technology, the application of intelligent unmanned systems in the military field will increase day by day. However, in the use of intelligent unmanned systems, autonomy and intelligence have not yet been fully realized. At present, the application status of intelligent unmanned system technology in the military field can be mainly divided into the following three parts:
① From the perspective of combat missions: combat missions have developed from simple reconnaissance and surveillance to mainstream confrontation operations; battlefield confrontation has changed from human confrontation to human-machine confrontation, and then to machine-machine confrontation; the application environment has changed from structured environment and laboratory environment to real battlefield environment, and will gradually develop into an augmented reality environment combining real environment and virtual reality in the future.
② From the perspective of command and control: the control method has developed from simple remote control and program control of a single machine to intelligent fusion and interactive control of human-machine, but autonomous control has not yet been fully realized; the system architecture has developed from specialization and singularity to generalization, standardization, and interoperability.
③ From the perspective of perception and decision-making: the decision-making method has changed from relying solely on people to relying mainly on people and supplemented by human-machine intelligent interactive decision-making; the perception method has changed from relying solely on sensors to obtain feature information and people to judge target attributes to target recognition and feature information acquisition based on artificial intelligence.
- Key technologies of intelligent unmanned systems
As a culmination of multidisciplinary fields, intelligent unmanned systems involve many technologies, perform diverse tasks, and have complex and changeable application scenarios. For example, the air environment is rainy and foggy, with low visibility, strong winds, and light interference; the land environment has complex terrain, obstacles, interference, and dangerous pollution areas; the sea environment has wind and wave interference, ship swaying, inconspicuous targets, and irregular coastlines. Different environments and uses pose huge challenges to the research and performance of intelligent unmanned system technology. In order to adapt to the restricted and changing environment, the key technologies of intelligent unmanned systems can be summarized as autonomous perception and understanding technology in complex environments, multi-scenario autonomous skill learning and intelligent control technology, multi-task cluster collaboration technology, human-computer interaction and human-computer fusion technology, decision-making planning technology and navigation and positioning technology. This section will mainly use marine unmanned systems as examples to elaborate on the key technologies of intelligent unmanned systems.
⒈ Autonomous perception and understanding technology in complex environments
Autonomous perception and scene understanding of the environment in complex environments is a prerequisite for intelligent unmanned systems to operate autonomously and form combat capabilities, which will directly affect whether the mission can be successfully completed. In view of the complexity and variability of the actual environment, especially the difficulties of wind and wave interference and ship shaking in the sea environment, intelligent unmanned systems need to complete the goals of autonomous target selection perception, obtain multimodal information, and abstract and complete understanding of information. Therefore, the autonomous perception and understanding technology of the environment of intelligent unmanned systems in complex environments needs to break through the autonomous perception technology of multimodal sensor fusion, as well as the complex scene target recognition and understanding technology.
⑴ Multimodal sensor fusion autonomous perception technology
At present, the information acquisition sensors carried by intelligent unmanned systems mainly include navigation radar, millimeter wave radar, laser radar, optoelectronic payload, etc. A single sensor cannot directly obtain high-precision, dense three-dimensional scene information. It is necessary to study the autonomous environmental perception technology of multi-sensor fusion to provide support for scene understanding. Multi-sensor fusion is to carry out multi-level and multi-space information complementation and optimization combination processing of various sensors, and finally produce a consistent interpretation of the observed environment. In this process, it is necessary to make full use of multi-source data for reasonable control and use, and the ultimate goal of information fusion is to derive more useful information based on the separated observation information obtained by each sensor through multi-level and multi-faceted combination of information. By taking advantage of the mutual cooperation of multiple sensors, the data of all information sources are comprehensively processed to improve the intelligence of the entire sensor system. The natural environment of the ocean is more complex than that of land and air. Faced with special challenges such as violent swaying of ships, wind and wave interference, uneven lighting, and inconspicuous targets, the marine intelligent unmanned system needs to perform multi-sensor information fusion processing on the designated target based on the unique attributes of each sensor, and then combine the electronic chart information of the internal navigation unit of the unmanned system and the shore-based support system to build a multi-dimensional three-dimensional situation map of the sea surface environment, perform tracking, detection, identification and cognition tasks for the designated target, and finally realize the autonomous perception and complete understanding of the sea surface environment by the marine intelligent unmanned system.
⑵ Complex scene target recognition and understanding technology
The key to the operation autonomy of intelligent unmanned systems lies in the ability to effectively understand the scene and target information, and accurate understanding of scene information mainly includes the construction of target semantic information and the description of scene text information. Compared with land and air environments, the natural marine environment faces unique difficulties such as wind and wave interference and violent swaying of the hull, which brings challenges to the intelligent unmanned system to fully understand the environmental information and accurately identify the designated target. Using sensors such as laser radar and high-definition cameras carried by intelligent unmanned systems, the original point cloud information and image feature information of the marine environment scene can be obtained. Using three-dimensional target detection methods based on point clouds, point clouds and image fusion, and three-dimensional scene semantic segmentation methods, etc., the intelligent unmanned system can fully recognize the scene information and accurately identify the designated target.
There are mainly two types of point cloud-based methods: grid-based or voxel-based methods, and point-based methods. The grid-based or voxel-based method uses voxels or bird’s-eye views to convert the irregular point cloud of the acquired sea surface into a regular representation method, and then extracts the point cloud features. The point-based method directly extracts target features from the acquired original point cloud of the sea surface. The three-dimensional target detection method based on point cloud and image fusion combines the precise coordinates of the target in the sea scene obtained by the laser radar with the environmental texture and color information provided by the sea surface image, which is more conducive to the intelligent unmanned system to accurately identify and accurately and completely understand the target of the ocean scene.
⒉ Behavior decision-making and trajectory planning technology
In actual and complex war scenes, for the complex mission environment and multiple tasks faced by intelligent unmanned systems, it is necessary to break through the behavior decision-making technology in multi-source heterogeneous environments, trajectory planning technology in dynamic/static environments, and trajectory tracking technology in complex scenes.
⑴ Behavior decision-making technology in multi-source heterogeneous environments
Behavior decision-making is the key to the realization of autonomous control of intelligent unmanned systems. In the complex environment of different speeds, different relative distances, and different data types of unmanned boats, it is necessary to accurately extract effective information to make safe and reliable control instructions for the next decision of the unmanned boat. First, extract representative environmental feature information and establish a sufficient number of accurately calibrated learning data sets; then, construct a decision maker based on a deep neural network and use the established database for learning; finally, use machine learning algorithms to optimize the constructed decision maker to further improve the decision accuracy.
⑵Trajectory planning technology in dynamic/static environment
Trajectory change is the most basic behavior of unmanned boats and unmanned submarines. In a complex battlefield environment, planning a feasible and reliable trajectory according to different environmental conditions is the key to the intelligent driving of unmanned boats and unmanned submarines. This technology mainly includes trajectory planning technology based on polynomials, trajectory planning technology based on multi-objective constraints, and trajectory planning technology based on positive and negative trapezoidal lateral acceleration.
⑶Trajectory tracking technology in complex scenes
Tracking the planned ideal trajectory is an important task for unmanned boats and unmanned submarines. The key lies in solving the problem of high-precision and high-stability control when unmanned boats or unmanned submarines track target trajectories. The main solution is: according to the kinematic and dynamic models of unmanned boats and unmanned submarines, the corresponding actuator control quantity is output to achieve real-time and accurate tracking of the specified target, and under the premise of ensuring tracking accuracy, the autonomous intelligent steering of unmanned boats and unmanned submarines and the coordinated control of multiple actuators of each drive module are realized.
⒊Autonomous navigation and positioning technology
The navigation and positioning system is a key component of the intelligent unmanned system, which can provide accurate and reliable information about the speed and position of unmanned boats or unmanned submarines. The navigation system is generally composed of gyroscopes, accelerometers, satellite receivers, etc., some of which are supplemented by visual modules, or are equipped with prior spatial position maps and physical information sensors based on actual complex environmental conditions. In order to achieve accurate execution of tasks, intelligent unmanned systems must break through navigation and positioning technology based on inertial/satellite deep information fusion, navigation and positioning technology based on inertial/astronomical information fusion, navigation technology based on visual tracking, and geophysical assisted navigation technology.
⑴ Navigation and positioning technology based on inertial/satellite deep information fusion
This technology introduces the inertial information of the unmanned boat into the satellite carrier/code loop, and then uses fully autonomous, short-term, and high-precision inertial information to assist the update of satellite receiver signals, thereby realizing the complementary advantages and optimal fusion of the inertial navigation and satellite navigation of the unmanned boat.
⑵ Navigation and positioning technology based on inertial/astronomical information fusion
The astronomical-based navigation system has the advantages of high autonomy and low susceptibility to interference. By using the information output by astronomical navigation and the information provided by the initial position, the position of the unmanned boat can be calculated. The fusion of inertial navigation information and astronomical navigation information can improve the robustness of astronomical navigation positioning. Inertial/astronomical combined positioning technology based on astronomical navigation assistance has become a key part of the field of autonomous navigation of unmanned systems.
⑶ Navigation technology based on visual tracking
Due to the complexity of the actual battlefield environment, unmanned boats will be in a complex working environment and are easily interfered by the outside world, resulting in GPS denial, which makes the navigation system unable to be in a combined state. A single inertial navigation system has low accuracy and is prone to accumulating errors. Long-term pure inertial navigation will make the unmanned boat lose the ability to perform tasks. However, the vision-based method does not have time error accumulation. It only needs to extract the key features of the image obtained by the high-definition camera to obtain the position information of the unmanned boat and the unmanned submersible through visual algorithms and prior knowledge. The vision-based navigation algorithm is not easily interfered with, has strong robustness, and can make up for the error accumulation caused by pure inertial navigation in a GPS denial environment, and is widely used.
⑷ Geophysical assisted navigation technology
Due to the unique environment of the ocean, unmanned submersibles need to sail underwater for a long time, resulting in the inability to obtain real-time and accurate satellite signals and astronomical information. In addition, due to problems such as weak underwater light, vision-based navigation methods are also limited. Therefore, by obtaining a priori spatial position map inside the ocean and using the field scene information obtained by the physical sensors carried by the unmanned submersible and matching them, high-precision autonomous navigation of the unmanned submersible can be achieved.
The temporal and spatial distribution characteristics of the inherent geophysical properties of the surveyed ocean can be used to produce a geophysical navigation spatial position map. By matching the physical feature information obtained by the physical property sensor carried by the unmanned submersible with the pre-carried spatial position map, the high-precision positioning of the unmanned submersible can be obtained, and the high-precision autonomous navigation of the unmanned submersible can be realized.
⒋ Multi-scenario autonomous skill learning and intelligent control technology
Multi-scenario intelligent control technology is a key technology for intelligent unmanned systems to solve complex, changeable and unstable control objects. It is an effective tool for intelligent unmanned systems to adapt to complex task requirements. In a complex marine environment, if intelligent unmanned systems want to complete real-time and accurate regional monitoring, target tracking, information acquisition and precision strikes, they must break through the autonomous skill learning technology of tasks, autonomous operation interactive control technology, and unmanned system motion control technology of human-like intelligent control.
⑴ Autonomous skill learning technology of tasks Autonomous
skill learning refers to the process of learning based on prior knowledge or rules to complete tasks in the process of interaction between unmanned systems and the outside world. The autonomous learning of unmanned system operation skills is essentially a partial process of simulating human learning cognition. Intelligent unmanned systems use deep reinforcement learning-based technology to combine the perception ability of deep learning with the decision-making ability of reinforcement learning, and can achieve direct control from high-latitude raw data information input to decision output in complex sea environments. The autonomous skill learning of intelligent unmanned systems mainly includes three aspects: first, describing the complex environment of the ocean surface and the interior of the ocean, and obtaining the initial state data information of the surrounding environment; second, based on the description of the intelligent unmanned system and the complex environment of the ocean surface and the interior, mathematical modeling of deep reinforcement learning is carried out to obtain key information such as the state value function and control strategy function of the autonomous skill learning process; third, using the data information obtained by the interaction between the intelligent unmanned system and the complex environment of the ocean surface and the interior, the state value function and the control strategy function are updated to enable the marine intelligent unmanned system to learn a better control strategy.
⑵ Autonomous operation interactive control technology
In the process of autonomous learning and control of tasks, the intelligent unmanned system needs to contact with the ocean surface and the complex internal environment to form a good coupling system to ensure the real-time and accurate acquisition of information on the ocean surface and the complex internal environment, and correctly and quickly carry out navigation planning, autonomous navigation control and autonomous collision avoidance of unmanned boats and unmanned submersibles. The tasks of the interactive control technology of autonomous operation of intelligent unmanned systems mainly include: the design of interactive rules and control strategies of intelligent unmanned systems; modeling methods of complex environments on the surface and inside of the ocean; online modeling and correction of the dynamics of unmanned boats, unmanned submarines and operating objects; dynamic generation and shared control methods of virtual force constraints in complex environments on the surface and inside of the ocean.
⑶ Motion control technology of unmanned systems with humanoid intelligent
control The motion control technology of unmanned systems with humanoid intelligent control combines artificial intelligence with traditional control methods to solve the problem of stable and precise control of unmanned boats and unmanned submarines in actual complex marine battlefield environments. It mainly includes two aspects: the design of intelligent control algorithms for unmanned systems and the design of intelligent control strategies for unmanned systems. The design of intelligent control algorithms for unmanned systems mainly includes: hierarchical information processing and decision-making mechanisms; online feature identification and feature memory; open/closed-loop control, positive/negative feedback control, and multi-modal control combining qualitative decision-making with quantitative control; the application of heuristic intuitive reasoning logic. The design of intelligent control strategies for unmanned systems is to design reasonable solutions for unmanned boats or unmanned submarines to meet actual mission requirements.
⒌ Unmanned cluster collaborative control technology
In actual combat scenarios, due to the complexity of the battlefield environment and the diversity of tasks, a single unmanned boat or unmanned submarine usually cannot meet the needs of actual tasks. The number of equipment carried by a single unmanned boat or unmanned submarine is limited, and the perception perspective and regional range are not comprehensive enough, resulting in insufficient precision and thoroughness in performing complete intelligence detection, target tracking, battlefield environment perception and comprehensive firepower strike tasks. Therefore, it has become an inevitable trend for a cluster of intelligent unmanned systems composed of multiple unmanned boats and unmanned submarines to collaboratively perform tasks. To complete the control of the intelligent unmanned system cluster, it is necessary to break through the local rule control technology of the intelligent unmanned system cluster, the soft control technology of the intelligent unmanned system cluster, the pilot control technology of the intelligent unmanned system cluster, and the artificial potential field control technology of the intelligent unmanned system.
⑴ Local rule control technology of intelligent unmanned system cluster
The control technology based on local rules is the basic method for intelligent unmanned systems to control unmanned boats and unmanned submarines. It mainly lies in the designation of individual local control rules within the cluster of unmanned boats and unmanned submarines. Local rule control technology has achieved intelligent control of marine unmanned system clusters to a certain extent, but a large number of experiments are needed to obtain the parameters between the behavior of marine unmanned system clusters and the cluster model, and the values of the parameters are also very sensitive. Therefore, to achieve complete intelligent control of intelligent unmanned systems, other technologies are needed.
⑵ Soft control technology of intelligent unmanned system clusters The
soft control technology of intelligent unmanned system clusters is mainly based on two requirements: First, in the intelligent unmanned system cluster, the control rules between individuals are very important. For example, the control and internal function of each unmanned boat and unmanned submarine are necessary conditions for the group behavior of the entire marine intelligent unmanned system cluster; second, the intelligent unmanned system cluster adopts a local communication strategy. With the increase of unmanned boats and unmanned submarines in the cluster system, it will not affect the state of the entire intelligent unmanned system cluster.
The soft control method is to add one or more new unmanned boats or unmanned submarines without destroying the individual rules of unmanned boats and unmanned submarines in the intelligent unmanned system cluster. These unmanned boats or unmanned submarines participate in the actions of the entire intelligent unmanned system cluster according to the same local rules, but they are controllable and can receive external instructions. After receiving the command, these unmanned boats or unmanned submarines will independently complete the corresponding tasks. The soft control method of the intelligent unmanned system cluster is to add a controllable unmanned boat and unmanned submarine on the basis of the local control rules of the unmanned system, so that it can affect the entire unmanned system cluster, and finally complete the control of the entire intelligent unmanned system group.
⑶ Intelligent unmanned system cluster navigation control technology
The basic content of the intelligent unmanned system cluster navigation control technology is: under the premise that the individuals of the entire marine intelligent unmanned system cluster maintain local rules, a small number of unmanned boats and unmanned submarines in the cluster have more information and stronger information processing capabilities, and interact with other unmanned boats and unmanned submarines through local information to play a leading role, so as to achieve the purpose of controlling the entire intelligent unmanned system cluster.
⑷ Artificial potential field control technology of intelligent unmanned system
In the control of intelligent unmanned system clusters, control technology based only on local rules is difficult to achieve accurate and real-time perception of the battlefield, as well as the collection and acquisition of intelligence information, tracking and identification of suspicious targets, and precise strikes on enemy areas. Artificial potential field control technology introduces the concept of potential field in physics into the control of intelligent unmanned system clusters, and uses potential functions to simulate the internal and external effects that affect a single unmanned boat or unmanned submarine. The single unmanned boat or unmanned submarine in the system cluster acts under the action of the potential function, and finally realizes the control of the entire intelligent unmanned system through the potential function.
⒍Natural human-computer interaction technology
In the actual battlefield environment, intelligent unmanned systems face problems such as complex operation tasks, low level of operation intelligence, high training risks and costs, and low equipment use and maintenance efficiency. In this case, it is necessary to improve the controllability and intelligence of intelligent unmanned system equipment, and it is necessary to break through the human-computer interaction technology of intelligent unmanned systems, augmented reality and mixed reality technology of intelligent unmanned systems, and brain-computer interface technology of intelligent unmanned systems.
⑴Human-computer interaction technology of intelligent unmanned systems
Human-computer interaction technology of intelligent unmanned systems refers to the command platform obtaining the image and voice information of officers and soldiers through image and voice sensors, and then using algorithms such as image segmentation, edge detection, and image recognition to extract key information such as gestures and eye gestures of officers and soldiers, and then using algorithms based on deep learning to obtain the voice information of officers and soldiers and pass it to the command platform, so as to issue the officers and soldiers’ instructions to lower-level combat units. The human-computer interaction technology of intelligent unmanned systems can improve the intelligence of task operations and the fault tolerance and robustness of the operation process, so that the officers and soldiers’ instructions can be issued to combat units more stably and effectively.
⑵Augmented reality and mixed reality technology of intelligent unmanned systems
Augmented reality technology of intelligent unmanned systems is to superimpose computer-generated images on real complex combat environments, and mixed reality technology of intelligent unmanned systems is to present information of virtual scenes in actual combat scenes, and set up an interactive feedback information loop between the virtual world and officers and soldiers in a real combat environment, thereby increasing the officers and soldiers’ sense of reality in the combat environment experience. As an important development direction of immersive human-computer interaction technology, virtual reality and augmented reality for intelligent unmanned systems have a variety of different real combat application scenarios, which can effectively reduce the cost and risk of training and improve the use and maintenance efficiency of equipment during combat.
⑶ Brain-computer interface technology for intelligent unmanned systems
The main function of the brain-computer interface is to capture a series of brain wave signals generated by the human brain when thinking. In actual combat environments, the brain-computer interface technology of intelligent unmanned systems extracts features and classifies the brain wave signals of commanders and fighters, thereby identifying the intentions of commanders and fighters and making corresponding decisions to cope with complex combat tasks and emergencies. The brain-computer interface technology of intelligent unmanned systems can enhance the cognitive and decision-making capabilities of commanders and fighters, greatly improve brain-computer interaction and brain control technology, and give commanders and fighters the ability to control multiple unmanned boats, unmanned submarines and other unmanned combat equipment while relying on thinking.
- Future development trend of intelligent unmanned systems
Due to its advantages of unmanned, autonomous, and intelligent, intelligent unmanned systems will appear in every corner of the future battlefield. As they undertake more battlefield tasks, they will participate in different war scenarios, which will lead to a number of key problems for intelligent unmanned systems, restricting their development. The key problems faced by intelligent unmanned systems are mainly:
① Highly complex environment. The specific application environment of intelligent unmanned systems will face more and more factors. The numerous shelters in unstructured environments, the limited perception viewpoints and ranges, etc., put forward higher requirements on the environmental perception ability of intelligent unmanned systems.
② High game confrontation. The battlefield game of intelligent unmanned systems is an important means to gain battlefield advantages. The fierce mobile confrontation between the two sides of the war, as well as the many interferences caused by the enemy and the battlefield environment, have put forward new challenges to the mobile decision-making ability of intelligent unmanned systems.
③ High real-time response. In the future battlefield, the combat situation will change dramatically, the combat mode will be more flexible and changeable, and it is necessary to respond to battlefield emergencies in a timely manner, which puts forward new requirements for the real-time response ability of intelligent unmanned systems.
④ Incomplete information. In the future battlefield, due to the limitations of the battlefield environment and the existence of enemy interference, the information acquisition ability of the intelligent unmanned system will be restricted, resulting in incomplete situational awareness, loss and attenuation of battlefield situation information data, and the inability to fully obtain information on both sides of the enemy.
⑤ Uncertain boundaries. The unmanned combat mode of the intelligent unmanned system has subverted the traditional combat mode. The integration of land, sea, air and space in the future unmanned combat, as well as the social public opinion brought about by the high degree of integration with society, will have an impact on the unmanned combat of the intelligent unmanned system, thus causing uncertainty in the combat boundary.
Based on the various difficulties that will be faced above, the development of intelligent unmanned systems in the future will focus on two aspects: individual capability enhancement and cluster capability enhancement. Individual capability enhancement is mainly reflected in individual cognitive intelligence, individual autonomous operation and algorithm chipization; cluster capability enhancement is mainly reflected in improving interoperability through a universal architecture, as well as cross-domain collaborative operations, network security and human-machine hybrid intelligence.
⒈ Cognitive intelligence adapts to complex task environments
In order to improve the adaptability of intelligent unmanned systems in highly complex environments, it is necessary to enhance the individual cognitive intelligence of intelligent unmanned systems. The enhancement of individual cognitive intelligence is mainly reflected in the transformation from individual perceptual intelligence to cognitive intelligence. The comprehensive acquisition of multi-source sensor information enables intelligent unmanned systems to have human semantic understanding, associative reasoning, judgment analysis, decision planning, emotional understanding and other capabilities. The development of individual cognitive intelligence of intelligent unmanned systems will be based on brain science and bionics, and will achieve intelligent understanding and accurate application of acquired information by combining knowledge graphs, artificial intelligence, knowledge reasoning, decision intelligence and other technologies, thereby improving the high real-time response capabilities of intelligent unmanned systems to emergencies.
⒉ Autonomous operation improves the task capability of single machines
In order to solve the problem of highly complex tasks faced by intelligent unmanned systems in highly complex environments, it is necessary to improve the autonomous operation capabilities of single machines. This includes developing decision-making methods based on deep reinforcement learning, autonomous environmental perception and interaction methods based on multi-source information of vision and other sensors, autonomous motion planning methods for robots based on neurodynamics, and autonomous operation methods based on artificial intelligence, so as to improve the autonomous environmental modeling and positioning capabilities, autonomous decision-making capabilities, autonomous planning capabilities and autonomous control capabilities of individuals in intelligent unmanned systems, so that intelligent unmanned systems can adapt to complex environments and carry out autonomous operation tasks.
⒊ Algorithm chipization achieves high real-time response
The complex environment faced by intelligent unmanned systems places high demands on algorithms and computing power. It is necessary to be able to accelerate computing in real time to achieve high real-time response to battlefield emergencies. To solve this problem, it is necessary to improve the chipization level of individual algorithms of intelligent unmanned systems, that is, to develop a new architecture of storage and computing integrated chips to improve the computing power of chips and the level of algorithm chipization. New chips based on artificial neural technology can be studied. By changing the binary computing method of digital chips and exchanging gradient signals or weight signals, the chips can work in a simulated neuron manner, simulating the parallel computing flow of the brain to effectively process large amounts of data, and obtaining the parallel computing capabilities of supercomputers, thereby greatly improving the computing power of chips and the level of algorithm chipization, and solving the problem of high real-time response of intelligent unmanned systems.
⒋ Universal architecture improves cluster interoperability
In order to improve the adaptability of intelligent unmanned systems facing highly complex environments and the maintenance and support efficiency of intelligent unmanned systems, intelligent unmanned systems will continue to develop standardized command and control frameworks in the future, improve the intelligence of human-machine collaboration, and improve the modularity of the system. It is mainly reflected in:
① Developing a general artificial intelligence framework to support autonomous, precise, and real-time good coupling and collaboration between humans and machines;
② Improving the modularity and component interchangeability of intelligent unmanned systems to support rapid maintenance and configuration upgrades of intelligent unmanned systems and their members in future battlefields;
③ Improving the level of data transmission integration and the anti-interference capability of data transmission on future battlefields to reduce the rate of data interception.
⒌ Cross-domain collaboration breaks the boundaries of cluster applications
In order to improve the adaptability of intelligent unmanned systems in highly complex environments and solve the problem of uncertain boundaries during combat, it is necessary to improve the cross-domain collaborative combat capabilities of intelligent unmanned systems to make up for the lack of capabilities in a single combat domain. Through the cross-domain collaborative combat of intelligent unmanned systems, the advantages of various components can be complemented. That is, by utilizing the advantages of large search range and long communication distance of air unmanned systems, as well as long endurance and strong stability of land unmanned systems and marine unmanned systems, the advantages of different components are combined to increase the multi-dimensional spatial information perception capabilities of intelligent unmanned systems, and form a heterogeneous multi-autonomous collaborative system, thereby improving the ability of intelligent unmanned systems to complete complex tasks.
⒍ Secure network guarantees reliable application of clusters
Intelligent unmanned systems face the problems of incomplete information and high game confrontation on future battlefields. Therefore, it is necessary to improve the network security protection capabilities of intelligent unmanned systems in high confrontation environments, improve flexibility in dealing with highly complex and highly variable tasks, and improve stability in the face of high-intensity network attacks. The improvement of network security protection capabilities in adversarial environments is mainly reflected in the following aspects:
① Plan reasonable data permissions to ensure data security and flexibility of task execution;
② Improve information protection capabilities, develop and upgrade information protection products for intelligent unmanned systems, and record response decisions for information explosion situations;
③ Increase the network’s deep defense capabilities, unify network security standards and levels, build network defense autonomy, and improve the network’s ability to resist attacks under network attacks.
⒎ Human-machine hybrid intelligence improves adversarial capabilities
In order to solve the problem of high real-time response faced on future battlefields and improve the adaptability of intelligent unmanned systems in highly complex environments, it is necessary to combine the advantages of humans and machines to form a new hybrid intelligent mode of human-machine collaboration, that is, to develop human-machine hybrid intelligence for intelligent unmanned systems. Human-machine hybrid intelligence of intelligent unmanned systems is a new intelligent scientific system that combines physics and biology in which human, machine, and environmental systems interact. In response to the problems of high-complexity environments and high real-time responses faced by intelligent unmanned systems on future battlefields, the development of human-machine hybrid intelligence in the future is mainly reflected in the following aspects:
① Information intelligence input. At the input end of information acquisition, the information data objectively collected by the sensors of the unmanned system equipment is combined with the subjective perception information of the combat commanders to form a multi-dimensional information acquisition and information input method.
② Intelligent information fusion. After obtaining multi-dimensional data information, a new data understanding method is constructed by integrating the computer’s calculation data with the information cognition of the combat commanders.
③ Intelligent information output. After the data information is fused and processed, the computer’s calculation results are matched with the value decisions of the combat commanders to form an organically combined probabilistic and regularized optimization judgment.
IV. Conclusion
Due to its autonomy, intelligence and unmanned characteristics, intelligent unmanned systems will play an increasingly important role in the future battlefield. The development of intelligent unmanned systems will also drive the development of intelligent computing, intelligent transportation, intelligent manufacturing, smart medical care, brain-like science and other disciplines. In the future, we should be guided by the mission requirements of actual complex battlefield environments, combine advanced technologies in cutting-edge disciplines such as artificial intelligence, and make overall top-level planning for intelligent unmanned systems; verify reliable airborne intelligent perception and intelligent computing equipment on different unmanned system combat platforms in land, air and marine unmanned systems, and develop reliable and stable key technologies such as unmanned system autonomous control, intelligent perception, intelligent decision-making and intelligent interaction, overcome the key difficulties of intelligent unmanned systems, and continuously improve the autonomous control, intelligent perception and intelligent decision-making capabilities of intelligent unmanned systems.
現代國語：
目前，無人系統裝備已在軍事衝突中嶄露頭角，例如，在土耳其與敘利亞的衝突中，土耳其利用空軍裝備的安卡-S型長航時無人機和巴拉克塔TB-2察打一體式無人機，對敘利亞政府軍進行了打擊；俄羅斯國防部也曾公佈敘利亞境內的武裝分子利用載有爆炸物的無人機對其軍事基地展開了集群式攻擊；2020年，美國利用一架MQ-9「收割者」無人機襲擊了伊朗高級軍事指揮官並使其當場斃命。無人作戰正在到來，智慧無人系統作為未來戰場的關鍵利器，將決定整個戰爭的勝利歸屬。
圖片來自網路
發展智慧無人系統不僅會推動現有軍事科技的升級與進步，還將帶動民用科技的智慧性發展，包括智慧交通系統、智慧家庭系統、智慧製造系統與智慧醫療系統等。為了更科學、快速地發展智慧無人系統，各科技大國紛紛推出了一系列有關智慧無人系統發展的規劃與路線，力求在智慧無人系統領域的發展中搶得先機，奪取制高點。相關的有美國的自主無人系統綜合路線圖、俄羅斯的國家武器裝備計畫、英國的國防創新技術框架、中國的新一代人工智慧發展計畫以及日本的中長期技術規劃等。
近年來，從空中到空間、從陸地到海洋，各種類型的智慧無人系統大量湧現，世界各國已經逐步將智慧無人系統部署到軍隊中，並且在一些地區衝突、反恐戰場中，智慧無人系統的關鍵作用日益增加。因此，本文將重點從未來戰場的軍事需求出發，基於未來戰場面臨的實際複雜環境的挑戰，分析智慧無人系統發展與應用所需的關鍵技術，並從軍事角度分析個體增強與集群增強關鍵技術，闡述智慧無人系統的發展趨勢。
一、國內外研究現狀
智慧無人系統概念才提出不久，目前其研究尚處於初級階段，國際上也未形成統一的定義，暫且將其定義為：由無人平台及若干輔助部分組成，具有感知、交互和學習能力，並且能夠基於知識進行自主推理、自主決策，從而達成目標的有機整體。智慧無人系統依據其作用的空間範圍，可劃分為陸地無人系統、空中無人系統和海洋無人系統三大部分。其中，陸地無人系統主要包括偵察無人車、運輸無人車、作戰無人車、破障無人車、排爆無人車、無人車編隊與指揮系統等；空中無人系統主要包括偵察無人機、作戰無人機、後勤運輸無人機以及無人機編隊等；海洋無人系統主要包括偵察無人艇、作戰無人艇、後勤運輸無人艇、巡邏搜救無人艇、偵察無人潛航器、作戰無人潛航器、岸基支援系統等。本節將從以上3個部分來對國內外智慧無人系統的研究現況進行闡述。
⒈國外智慧無人系統研究現狀
⑴陸地無人系統
陸地無人系統主要用於情報蒐集、偵察巡邏、掃雷除障、火力打擊、戰場救援、後勤運輸、通信中繼以及電子乾擾等領域，隨著陸地無人系統在戰鬥中的優勢愈發凸顯，針對其的研究愈發受到各國的廣泛關注。
美國曾於1993年11月啟動「聯合戰術無人車」項目，亦即「角鬥士」無人作戰平台項目的前身。 2006年，美國完成了「角鬥士」無人作戰平台全系統的設計，並於2007年正式裝備海軍陸戰隊。 「角鬥士」戰術無人作戰平台是世界上第1款多用途作戰無人平台，搭載的感測器系統有日/夜攝影機、GPS定位系統以及聲學與雷射搜尋系統等，並裝備有機槍、衝鋒槍、催淚彈、狙擊手系統、生化武器探測系統等，可以在不同的天氣和地形下執行偵察、催淚彈、狙擊手電擊
「角鬥士」無人作戰平台搭載有高機動與高生存底盤，針對該平台，還開發了便攜式手持控制系統，並圍繞該控制系統的抗干擾性、網絡互操作性、小型化與操縱簡便化等技術問題完成了一系列開發工作。但因「角鬥士」無人作戰平台的裝甲防護能力較弱，執行任務的隱蔽性差，導致其遠程偵察與控制系統面臨的干擾較多。除此之外，美國陸軍還服役了一些其他的陸地無人系統，如「蝎子」機器人、「魔爪」機器人等。 2017年，美國陸軍制定了《機器人與自主系統（RAS）戰略》，為進行無人作戰能力建構提供了頂層規劃。圖1所示為美國陸地無人系統。
圖1 美國陸地無人系統
以色列、俄羅斯、英國和德國也相繼進行了陸地無人系統的研發工作，並研發出了一系列先進的產品，產品清單如表1所示。例如，以色列研發的「守護者」系列自主無人車可以結合搭載的傳感器與融合演算法，自主偵察與識別危險障礙，執行巡邏、監視與小規模的火力打擊任務；俄羅斯研製的MARSA-800無人車可以執行運輸和後勤保障障礙以及跟踪監視等任務，並可以在執行任務的過程中實現自主路徑規劃，規避障礙，該程序已部署。英國和德國對陸地無人系統的研究也開展得較早，英國於上世紀60年代就推出了手推車排爆機器人，後來又推出HarrisT7觸覺反饋機器人，用於執行拆彈、排爆等危險任務；德國萊茵金屬公司開發的「任務大師」地面武裝偵察無人車主要用於執行戰術監視、危險物品；德國萊茵金屬公司開發的「任務大師」地面武裝偵察無人車輛主要用於執行戰術監視、危險物品檢測、醫療後送機、消防系統
表1 各國陸地無人系統
⑵空中無人系統
空中無人系統主要以單一無人機平台和無人機集群為主。無人機由於具有視野開闊、飛行自由、設備搭載性好等優點，被廣泛應用於軍事領域，並在近年來的軍事衝突中發揮了極大的作用。空中無人系統的主要功能包括：情報蒐集、偵察監視、誘餌靶機、目標追蹤、戰術打擊與空中救援等。
美國空軍研究實驗室於2000年提出了針對無人機自主作戰的概念，並對無人機的自主程度進行了量化定義，並制定了發展計畫。無人機自主程度量化內容與發展階段如圖2所示。
圖2 自主控制水準與無人機自主化趨勢
2003年，美國將空軍和海軍的無人作戰飛機系統項目合併，啟動了「聯合無人作戰系統」（J-UCAS）項目，開始了對無人作戰飛機X-47B的研究。 2006年，美海軍提出了「海軍無人作戰航空系統」(N-UCAS)項目，旨在為航空母艦載機聯隊引入無人作戰飛機，並繼續對X-47B開展研究。在2012—2014年間，又多次完成了航母彈射、著艦、觸艦復飛等試驗，並於2015年完成了自主空中加油試驗。 X-47B攻擊型無人機是一款可以自主操縱、隱身性能好且適用於陸基和艦載的無人作戰飛機，具備高航程和高航時的特點，裝備有照射雷達、光電導引系統和孔徑雷達等先進的感測器，主要功能包括情報偵察、目標追蹤、電子戰幹擾、火力打擊等。美國研發的其他空中無人系統，如「全球鷹」、「掠食者」、「獵人」和「大烏鴉」等也已在軍隊服役，如圖3所示。
以色列研發的「哈比」無人機配備反雷達感應器、光電導引系統和飛彈，可自主攻擊敵方雷達系統，如圖3所示。
圖3 各國空中無人系統
單一空中無人系統在執行任務時容易被幹擾和打擊從而導致任務失敗，而空中無人系統集群則可以彌補這一缺陷，更大程度地發揮空中無人系統的優勢。美國國防先進研究計畫局（DARPA）針對空中無人系統集群先後啟動了「小精靈」低成本無人機計畫、低成本無人機集群計畫、「山銻」（Perdix）微型無人機機載高速發射展示項目、進攻性蜂群使能戰術（OFFSET）項目等，透過開發和測試用於無人系統集群的體系架構、通訊系統以及分散式控制演算法，發展了無人機集群自主控制系統，並利用人工智慧、態勢感知、虛擬實境和擴增實境等前沿科學技術，提升了空中無人系統集群在戰場上的綜合作戰能力。
⑶海洋無人系統
海洋無人系統包括水面無人系統及水下無人系統2類。其中，水面無人系統主要指水面無人艇（以下簡稱「無人艇」），主要用於執行海上搜救、偵察監視、火力打擊、巡邏安防、電子乾擾、後勤保障及誘餌靶船等任務；水下無人系統主要指無人潛航器，與執行人潛艦相比，其具無性戰力戰、高防震力與高威力控制權。 2018年，美海軍發布了《海軍部無人系統戰略路線圖》，2019年，又發布了《海軍人工智慧框架》，為海軍作戰與海洋無人系統的發展提供了路線規劃與指南。
在水面無人系統方面，美國提出了「美國先進概念技術演示計畫」（ACTD），其重要任務之一便是開展「斯巴達偵察兵」無人艇的研究。該計畫已於2007年完成，並在伊拉克戰區進行了試驗。 「斯巴達偵察兵」無人艇搭載有無人駕駛系統與視距/超視距通訊系統，並搭載有電光/紅外線搜尋轉塔、高畫質攝影機、導航雷達、水面搜索雷達、全球定位系統接收機等先進感測器，以及艦砲、反艦飛彈及反潛感應器等武器，主要用於執行情報蒐集、具有防監視、情報、反艦飛彈及反潛感應器等武器，主要用於執行情報蒐集、具有防監視、情報、反艦導彈及反潛感應器等武器，主要用於執行情報蒐集、具有防監視、情報、反艦導彈及反潛感美國研發的「海上獵人」無人艇搭載有聲吶與光電感測器，以及近距、遠程雷達偵測系統與可擴展模組化聲吶系統，主要用於執行辨識、監測可疑目標，引導火力打擊等任務。美國海洋無人系統如圖4所示。以色列研發的「保護者」無人艇主要用於執行情報偵察、可疑目標辨別、戰術攔截、電子乾擾和精確打擊等任務（圖4）。俄羅斯研發的無人水面偵察艇可以在母艦的指揮下執行快速巡邏任務並檢查、監視指定區域，搜尋情報。
圖4 各國海洋無人系統
在水下無人系統方面，俄羅斯開發的核動力無人潛航器“波塞冬”，可攜帶常規以及核彈頭，執行偵察與戰略核打擊任務，如圖4所示。美國研發的「刀魚」無人潛航器，可透過發出低頻電磁波來掃描可疑物體，搜尋情報；研發的「鮪魚」-9無人潛航器可攜帶多種標準載重，可用於執行近海勘探、反水雷、監視和偵察（ISR）等任務。
⒉國內智慧無人系統研究現狀
近年來，我國軍用智慧無人系統發展迅速，本文將從陸地無人系統、空中無人系統和海洋無人系統3個面向進行闡述。
在陸地無人系統方面，國防科技大學與三一重工股份有限公司共同開發了「沙漠蒼狼」陸地無人輕型平台，其以履帶為動力，搭載榴彈發射器和機槍等武器系統，可以用來執行後勤運輸、傷員運送、偵察監測、火力打擊等任務。山河智慧集團開發的「龍馬」系列無人車，具有強大的運輸與越障能力。南京理工大學研發的「神行-III」軍用地面智慧機器人系統，具有較強的自主導航與情報偵察能力。國防科技大學與哈爾濱工業大學等單位聯合研發的無人駕駛核化偵察車，具有較高的機動能力與裝甲防護能力，搭載的武器系統可以執行火力打擊並具備一定的自主能力。
在空中無人系統方面，成都飛機工業集團開發的「翼龍」系列無人機具有全自主水平起降能力、巡航飛行能力、空地協同能力與地面接力控制能力等，搭載有多型光電/電子偵察設備以及小型空地精確打擊武器，可以執行情報偵察、目標跟踪、火力打擊等任務。我國研發的「彩虹」系列無人機具有中空長航時的航行能力，可搭載電子乾擾系統與多種武器系統，能執行火力打擊、情報偵察、通訊幹擾、電波幹擾等任務；研發的攻擊11型無人機具有極強的隱身能力，可搭載精確的導引飛彈，用於執行對地導攻擊任務。我國空中無人系統如圖5所示。
圖5 我國空中無人系統
在海洋無人系統的水面無人系統方面，由哈爾濱工程大學主導開發的「天行一號」無人艇，採用油電混合動力，最高航速超過92.6km/h，最大航程1000km，為目前世界上最快的無人艇。該艇融合了自主感知、智慧控制、自主決策等技術，可實現對周圍複雜環境的快速態勢資訊認知與危險規避，可用於執行氣象資訊監控、地形測繪、警戒巡邏、情報偵察、火力攻擊等任務。由上海大學研發的「精海」系列無人艇具有半自主與全自主的作業能力，可執行目標偵察、海洋測繪、水質檢測等任務。由上海海事大學研發的「海騰01」號智慧高速無人艇，搭載有毫米波雷達、雷射雷達、前視聲吶等感測器，可執行可疑目標監視、水下測量、海上搜救等任務，具備全自主與半自主航行能力。江蘇自動化研究所研發的JARI智慧無人作戰艇，搭載有光電偵測器、四面相控陣等偵測設備，同時，也搭載有飛彈魚雷等武器系統，可以執行情報蒐集、敵情偵察、精準火力打擊等任務。由珠海雲洲智慧科技有限公司等單位聯合研發的「瞭望者Ⅱ」無人飛彈艇，搭載全自主無人駕駛系統及飛彈等武器，可執行敵情偵察、情報蒐集、精準火力打擊等任務。我國海洋無人系統如圖6所示。
圖6 我國海洋無人系統
在海洋無人系統的水下無人系統方面，西北工業大學開發的「魔鬼魚」無人潛航器為仿生蝠鱝無人潛水器，已完成了1025m的深海測試。由哈爾濱工程大學研發的「悟空號」全海深無人潛航器，成功完成了10896m的深潛和自主作業試驗。我國研發的「潛龍一號」、「海馬號」等深海潛水器都已成功完成深海探測任務。
⒊技術現況總結
目前，智慧無人系統已逐步應用於軍事應用的各個領域，隨著前沿科學技術的發展，智慧無人系統在軍事領域的應用將日益增加。但在智慧無人系統的使用方面，尚未完全實現自主化與智慧化。目前，智慧無人系統技術在軍事領域的應用現況主要分為以下3個部分：
①從作戰任務的角度：作戰任務從執行簡單的偵察監視向主流對抗作戰方向發展；戰場對抗由人人對抗向人機對抗，再向機機對抗方式轉變；應用環境由結構化環境、實驗室環境向真實戰場環境轉變，並在未來逐步發展成真實環境與虛擬現實相結合的增強現實環境。
②從指揮控制的角度：控制方式從單機簡單遙控、程控方式向人機智慧融合互動控制方向發展，不過尚未完全實現自主控制；體系結構由專用化、單一化向通用化、標準化、互通性方向發展。
③從感知決策的角度：決策方式由單一依靠人來決策向以人為主，人機智能交互決策為輔的方式轉變；感知方式由單一依靠傳感器獲取特徵信息，由人來判斷目標屬性向基於人工智能的目標識別、特徵信息獲取的方式轉變。
二、智慧無人系統關鍵技術
智慧無人系統作為多學科領域的集大成者，涉及的技術眾多，執行的任務多樣，且應用場景複雜多變。例如，空中環境多雨、多霧，能見度低，有大風、光照幹擾等；陸地環境地形複雜，有障礙物遮擋幹擾和危險污染區域等；海上環境有風浪幹擾、船舶搖擺、目標不顯著、海岸線不規則等。不同的環境及用途給智慧無人系統技術研究和性能的發揮提出了巨大挑戰。為適應受限的多變環境，可將智慧無人系統關鍵技術歸納為複雜環境下自主感知與理解技術、多場景自主技能學習與智慧控制技術、多任務集群協同技術、人機互動與人機融合技術、決策規劃技術與導航定位技術，本節將主要以海洋無人系統為案例對智慧無人系統關鍵技術進行詳細闡述。
⒈複雜環境下自主感知與理解技術
在複雜環境下對環境進行自主感知與場景理解是智慧無人系統能夠自主作業並形成作戰能力的前提，將直接影響任務能否成功完成。針對實際環境的複雜多變，尤其是海面環境的風浪幹擾及船舶搖晃等困難，智慧無人系統需要完成目標自主選擇感知，獲取多模態訊息，並對資訊抽象完整理解等目標。因此，複雜環境下的智慧無人系統環境自主感知與理解技術需突破多模態感測器融合自主感知技術，以及複雜場景目標辨識與理解技術。
⑴多模態感測融合自主感知技術
目前，智慧無人系統搭載的資訊取得感測器主要包括導航雷達、毫米波雷達、光達、光電載重等。單一感測器無法直接獲取高精度、稠密的場景三維訊息，需研究多感測器融合的環境自主感知技術，從而為場景理解提供支撐。多感測器融合是將各種感測器進行多層次、多空間的資訊互補和最佳化組合處理，最終產生對觀測環境的一致性解釋。在此過程中，要充分利用多源數據進行合理的支配與使用，而信息融合的最終目標則是基於各傳感器獲得的分離觀測信息，通過對信息多級別、多方面組合導出更多有用的信息。透過利用多個感測器相互協同操作的優勢，綜合處理所有資訊來源的數據，從而提高整個感測器系統的智慧化。海洋自然環境相比陸地與空中環境更為複雜，面臨船舶的劇烈搖擺、風浪幹擾、光照不均、目標不顯著等特殊的挑戰，海洋智慧無人系統需要依據每種感測器的獨特屬性來對指定目標進行多感測器資訊融合處理，接著結合無人系統內部導航單元與岸基支援系統的電子海圖訊息，建構海面環境多維立體態勢圖，執行對指定目標的追蹤、偵測、辨識與認知任務，最終實現海洋智慧無人系統對海面環境的自主感知與完整理解。
⑵複雜場景目標辨識與理解技術
智慧無人系統具備作業自主性的關鍵在於能有效理解場景與目標訊息，而準確理解場景資訊主要包括目標語意訊息建構與場景文字訊息描述。相較於陸地與空中環境，海洋自然環境面臨風浪幹擾、船體劇烈搖擺等獨特的困難，這為智慧無人系統完整地理解環境資訊與準確識別指定目標帶來了挑戰。利用智慧無人系統搭載的雷射雷達與高清攝影機等感測器，可以獲得海洋環境場景的原始點雲信息及影像特徵信息，利用基於點雲、點雲與影像融合的三維目標檢測方法與三維場景語義分割方法等，可以實現智慧無人系統對場景資訊的完整認知及對指定目標的準確識別。
基於點雲的方法主要包括2種：基於網格或體素的方法，以及基於點的方法。基於網格或體素的方法是利用體素或鳥瞰圖來將所獲得的海面不規則的點雲轉換成規則的表徵方式，然後提取點雲特徵。基於點的方法則是直接在所獲取的海面原始點雲中提取目標特徵。基於點雲與影像融合的三維目標檢測方法，是將雷射雷達獲得的海面場景中目標的精確座標與海面影像提供的環境紋理和色彩資訊相結合，這樣更加有助於智慧無人系統對海洋場景目標的精確識別與準確、完整的理解。
⒉行為決策與軌跡規劃技術
在實際的、複雜的戰爭場景中，對於智慧無人系統面臨的複雜任務環境與多重任務，必須突破多源異質環境下的行為決策技術、動/靜環境下的軌跡規劃技術與複雜場景下的軌跡追蹤技術。
⑴多源異質環境下的行為決策技術
行為決策是智慧無人系統實現自主控制的關鍵。在無人艇不同速度、不同相對距離、不同資料類型的複雜環境下，需要準確提取有效資訊來為無人艇下一刻的決策做出安全可靠的控制指令。首先，提取出具有代表性的環境特徵信息，建立足夠數量與精確標定的學習數據集；然後，構建基於深度神經網絡的決策器，並利用建立的數據庫進行學習；最後，利用機器學習算法對構建的決策器進行優化，進一步提高決策精度。
⑵動/靜環境下的軌跡規劃技術
軌跡變換是無人艇與無人潛航器最基本的行為。在複雜的戰場環境下，根據不同的環境狀況規劃一條可行、可靠的軌跡是無人艇與無人潛航器實現智慧行駛的關鍵。此技術主要包括基於多項式的軌跡規劃技術、基於多目標限制的軌跡規劃技術與基於正、反梯形側向加速度的軌跡規劃技術。
⑶複雜場景下的軌跡追蹤技術
對規劃出的理想軌跡進行追蹤是無人艇與無人潛航器的重要任務，其關鍵在於解決無人艇或無人潛航器進行目標軌跡追蹤時的高精度與高穩定性控制難題。主要解決方法為：根據無人艇與無人潛航器的運動學與動力學模型，輸出對應的執行器控制量來實現對指定目標的即時、準確跟隨，在保證追蹤精度的前提下，實現無人艇與無人潛航器的自主智慧轉向與各個驅動模組多執行器之間的協調控制。
⒊自主導航定位技術
導航定位系統是智慧無人系統的關鍵組成部分，其可提供精準、可靠的有關無人艇或無人潛航器的速度與位置等資訊。導航系統一般由陀螺儀、加速計、衛星接收器等組成，部分輔以視覺模組，或基於實際複雜的環境狀況搭載先驗空間位置圖與實體資訊感測器等。智慧無人系統要實現任務的精準執行，必須突破基於慣性/衛星深度資訊融合導航定位技術、基於慣性/天文資訊融合導航定位技術、基於視覺追蹤的導航技術與地球物理輔助導航技術。
⑴基於慣性/衛星深度資訊融合的導航定位技術
該技術是將無人艇的慣性資訊引入衛星載波/碼環路，然後利用全自主、短時、高精度的慣性資訊輔助衛星接收機訊號的更新，從而實現無人艇的慣性導航與衛星導航的優勢互補及最適融合。
⑵基於慣性/天文學資訊融合的導航定位技術
基於天文的導航系統具有高自主性與不易受干擾的優勢，透過利用天文導航輸出的信息與初始位置提供的信息，可以推算出無人艇的位置。將慣性導航資訊與天文導航資訊融合，可以提高天文導航定位的穩健性。基於天文導航輔助的慣性/天文組合定位技術已成為無人系統自主導航領域的關鍵部分。
⑶基於視覺追蹤的導航技術
由於實際戰場環境的複雜性，無人艇會處於複雜的工作環境中，容易受到外界幹擾而出現GPS拒止的情況，使導航系統無法處於組合狀態。單獨的慣性導航系統精度較低，容易累積誤差，長時間的純慣性導航會使無人艇失去執行任務的能力。而基於視覺的方法卻沒有時間的誤差積累，只需提取到高清相機所獲得影像的關鍵特徵，即可透過視覺演算法與先驗知識獲得無人艇與無人潛航器的位置資訊。基於視覺的導航演算法不易受到干擾，魯棒性較強，且能彌補在GPS拒止環境下由純慣性導航帶來的誤差積累，被廣泛應用。
⑷地球物理輔助導航技術
由於海洋獨特的環境，無人潛航器需長時間在水下航行，導致無法取得即時、準確的衛星訊號與天文資訊。另外，由於水下光照弱等問題，基於視覺的導航方法也受到限制。因此，透過獲得海洋內部的先驗空間位置圖，並利用無人潛航器搭載的物理感測器所獲得的實地場景資訊並進行匹配，可以實現無人潛航器的高精度自主導航。
可以利用勘測的海洋固有的地球物理屬性的時空分佈特徵，來製作地球物理導航空間位置圖，透過將無人潛航器所搭載的物理屬性感測器實地獲取的物理特徵資訊與預先搭載的空間位置圖相匹配，可以獲得無人潛航器的高精度定位，實現無人潛航器的高精度自主導航。
⒋多場景自主技能學習與智慧控制技術
多場景智慧控制技術是智慧無人系統解決複雜、多變和控制物件不穩定等問題的關鍵技術，是智慧無人系統適應複雜任務需求的有效工具。在複雜的海洋環境下，智慧無人系統要完成即時、準確的區域監控、目標追蹤、資訊取得與精準打擊，就必須突破任務的自主技能學習技術、自主作業互動控制技術，以及類人智慧控制的無人系統運動控制技術。
⑴任務的自主技能學習技術
自主技能學習是指在無人系統與外界互動的過程中，基於先驗知識或規則進行學習以完成任務的過程。無人系統作業技能的自主學習本質是模擬人學習認知的部分過程。智慧無人系統利用基於深度強化學習的技術，將深度學習的感知能力與強化學習的決策能力相結合，可實現在海面複雜環境下從高緯度的原始資料資訊輸入到決策輸出的直接控制。智慧無人系統自主技能學習主要包括3個面向：一是對海洋表面與海洋內部的複雜環境進行描述，並獲得周圍環境的初始狀態資料資訊；二是基於智慧無人系統與海洋表面和內部複雜環境的描述方式，進行深度強化學習的數學建模，獲得自主技能學習過程的狀態價值函數與控制策略函數等關鍵信息；三是利用智能無人系統與海洋表面和內部複雜環境交互所獲得的數據信息，對狀態價值函數及控制策略函數進行更新，以使海洋智能無人系統學習出更優的控制策略。
⑵自主作業互動控制技術
智慧無人系統在任務的自主學習與控制過程中，需要與海洋表面和內部複雜環境接觸形成良好的耦合系統，以確保對海洋表面與內部複雜環境資訊的即時、準確獲取，並正確、快速進行無人艇、無人潛航器的航行規劃、自主航行控制與自主規避碰撞等。智慧無人系統自主作業互動控制技術的任務主要包括：智慧無人系統互動規則與控制策略的設計；海洋表面與內部複雜環境的建模方法；無人艇、無人潛航器與作業物件的動力學線上建模及修正；海洋表面與內部複雜環境中虛擬力約束的動態生成及共享控制方法。
⑶類人智慧控制的無人系統運動控制技術
類人智慧控制的無人系統運動控制技術是將人工智慧與傳統控制方法結合，以解決在實際複雜的海洋戰場環境下，無人艇與無人潛航器的穩定精確控制問題，主要包括無人系統智慧控制演算法的設計與無人系統智慧控制策略的設計2個面向。無人系統智慧控制演算法設計主要包括：分層的資訊處理和決策機構；線上的特徵辨識與特徵記憶；開/閉環控制、正/負回饋控制以及定性決策與定量控制相結合的多模態控制；啟發式直覺推理邏輯的運用。無人系統智慧控制策略設計則是設計合理的無人艇或是無人潛航器的方案，以滿足實際的任務需求。
⒌無人群聚協同控制技術
在實際的作戰場景中，由於戰場環境的複雜性與任務的多樣性，單艘無人艇或是無人潛航器通常都無法滿足實際任務的需求。單艘無人艇或無人潛航器搭載的設備數量有限，感知視角與區域範圍不夠全面，導致在執行完整的情報探測、目標跟踪、戰場環境感知與全面火力打擊任務時不夠精確與徹底，因此，由多艘無人艇與無人潛航器組成的智能無人系統集群協同執行任務就成為必然的趨勢。要完成對智慧無人系統集群的控制，需要突破智慧無人系統集群局部規則控制技術、智慧無人系統集群軟控制技術、智慧無人系統集群領航控制技術以及智慧無人系統人工勢場控制技術。
⑴智慧無人系統叢集局部規則控制技術
基於局部規則的控制技術是智慧無人系統針對無人艇、無人潛航器集群控制的基本方法，主要在於對無人艇、無人潛航器集群內部個體局部控制規則的指定。局部規則控制技術在一定程度上實現了對海洋無人系統集群的智慧控制，但是對於海洋無人系統集群行為與集群模型之間的參數，需要進行大量的實驗來獲得，並且對參數的取值也非常敏感。所以，要實現對智慧無人系統完全的智慧控制，還需輔助以其他技術。
⑵智慧無人系統叢集軟控制技術
智慧無人系統集群的軟控制技術主要基於2點需求：一是在智慧無人系統集群中，個體之間的控制規則很重要，例如每艘無人艇、無人潛航器的控制與內部作用是整個海洋智慧無人系統集群出現群體行為的必要條件；二是智慧無人能動工具的控制與內部作用是整個海洋智慧無人系統集群出現群體行為的必要條件；二是智慧無人能動系統採用的是局部通訊策略，隨著智慧客系統集群出現群體行為的必要條件）
軟控制方法是在不破壞智慧無人系統集群內部無人艇、無人潛航器個體規則的前提下，加入一個或多個新的無人艇或是無人潛航器，這些無人艇或無人潛航器按照同樣的局部規則來參與整個智能無人系統集群的行動，但本身可控，可以接收外部指令。在接收指令後，這些無人艇或無人潛航器將獨立完成相應的任務。智慧無人系統集群的軟控制方法是在無人系統局部控制規則的基礎上，加入一個可以控制的無人艇與無人潛航器，使其對整個無人系統集群產生影響，最終完成對整個智慧無人系統群體的控制。
⑶智慧無人系統叢集領航控制技術
智慧無人系統集群領航控制技術的基本內容是：在整個海洋智慧無人系統集群個體保持局部規則的前提下，令集群中少數無人艇與無人潛航器擁有更多的信息量和更強的信息處理能力，並與其他無人艇和無人潛航器通過局部信息交互來起到領導者的作用，從而達到控制整個智能沒有集群的目的。
⑷智慧無人系統人工勢場控制技術
在智慧無人系統集群控制中，只基於局部規則的控制技術難以完成對戰場準確、即時的感知，以及對情報資訊的蒐集獲取、對可疑目標的追蹤識別和對敵方區域的精準打擊。人工勢場控制技術是將物理學中的位能場概念引入智慧無人系統集群的控制中，利用位勢函數來模擬影響單艘無人艇或無人潛航器的內、外作用，而係統集群中的單艘無人艇或無人潛航器則在勢函數的作用下行動，最終透過勢函數來實現對整個智慧無人能動系統的控制。
⒍自然人機互動技術
在實際的戰場環境中，智慧無人系統面臨著操作任務複雜、操作智慧化程度低、訓練風險大且成本高、設備使用與維修效率低等問題，在這種情況下，就需要提高智慧無人系統設備的可操控性與智慧化，需要突破智慧無人系統人機互動技術、智慧無人系統擴增實境與混合實境技術以及智慧無人系統介面技術。
⑴智慧無人系統人機互動技術
智慧無人系統人機互動技術是指指揮平台透過影像和語音感應器獲取指戰員的影像與語音訊息，然後利用影像分割、邊緣偵測、影像辨識等演算法擷取出指戰員的手勢與眼勢等關鍵訊息，接著利用基於深度學習的演算法獲得指戰員的語音訊息並傳遞給指揮平台，從而將指作戰員的指令下發給下級的指令。智慧無人系統的人機互動技術可以提高任務操作的智慧化以及操作過程的容錯率與魯棒性，從而使指戰員的指令能夠更加穩定、有效地下發給作戰單位。
⑵智慧無人系統擴增實境與混合實境技術
智慧無人系統擴增實境技術是將電腦生成的影像疊加在真實的複雜作戰環境中，智慧無人系統混合實境技術則是透過在實際作戰場景中呈現虛擬場景的訊息，在真實的作戰環境下在虛擬世界與指戰員之間搭起一個互動回饋的資訊迴路，從而增加指戰員對作戰環境體驗的真實感。智慧無人系統虛擬實境與擴增實境作為沉浸式人機互動技術的重要發展方向，已有多種不同的真實作戰應用場景，可有效降低訓練時的成本與風險，提高作戰時設備的使用與維修效率。
⑶智慧無人系統腦機介面技術
腦機介面的主要功能是捕捉人腦在進行思考活動時產生的一系列腦波訊號。在實際作戰環境中，智慧無人系統腦機介面技術透過對指戰員的腦波訊號進行特徵提取、功能分類，從而辨別出指戰員的意圖而做出相應的決策，以此應對複雜的作戰任務與突發情況。智慧無人系統腦機介面技術可以增強指戰員的認知與決策能力，大幅提升腦機互動與腦控技術，賦予指戰員在藉助思維的同時具有能操控多艘無人艇與無人潛航器等無人作戰設備的能力。
三、智慧無人系統未來的發展趨勢
智慧無人系統由於其無人化、自主性、智慧性等優點，將出現在未來戰場的各個角落，而隨著其承擔戰場任務的增多，將會參與不同的戰爭場景，導致智慧無人系統將面臨多項關鍵性的難題，使其發展受到限制。智慧無人系統面臨的關鍵性難題主要有：
①環境高度複雜。智慧無人系統具體的應用環境將面臨越來越多的要素，非結構化環境下遮蔽物眾多、感知視點及範圍受限等對智慧無人系統的環境感知能力提出了更高的要求。
②博弈高對抗。智慧無人系統的戰場博弈是取得戰場優勢的重要手段，作戰雙方激烈的機動對抗，以及因敵方和戰場環境帶來的諸多幹擾對智慧無人系統的機動決策能力提出了新的挑戰。
③響應高實時。在未來戰場中，戰鬥態勢變化劇烈，交戰方式將更加靈活多變，需及時應對戰場突發事件，這就對智慧無人系統的即時響應能力提出了新的要求。
④資訊不完整。在未來戰場中，受戰場環境的限制以及敵方幹擾的存在，智慧無人系統的資訊取得能力將會受到製約，從而造成態勢感知不完備、戰場態勢資訊資料遺失與衰減，導致無法完整取得敵我雙方的資訊。
⑤邊界不確定。智慧無人系統的無人作戰方式顛覆了傳統作戰模式，未來無人作戰的陸海空天一體化，以及透過與社會高度交融帶來的社會輿情，都將對智慧無人系統的無人作戰產生影響，從而造成作戰邊界的不確定性。
基於以上將面臨的各種難題，未來智慧無人系統的發展將集中在個體能力增強與群聚能力增強2個面向。個體能力增強主要體現在個體認知智能、個體自主作業與演算法晶片化等方面；集群能力增強則主要體現在透過通用化架構提升互通性，以及跨域協同作戰、網路安全與人機混合智能等。
⒈認知智能適應複雜任務環境
為提高智慧無人系統在高度複雜環境下的適應能力，需要增強智慧無人系統的個別認知智能。個體認知智能增強主要體現在從個體感知智能轉變為認知智能的轉變方面，綜合獲取的多源感測資訊使得智能無人系統具備人類的語意理解、聯想推理、判斷分析、決策規劃、情感理解等能力。智慧無人系統個體認知智能的發展將以腦科學和仿生學等為基礎，透過結合知識圖譜、人工智慧、知識推理、決策智慧等技術來實現獲取資訊的智慧理解與準確運用，從而提升智慧無人系統對突發事件的高即時響應能力。
⒉自主作業提升單機任務能力
為解決智慧無人系統在高度複雜環境下所面臨的高度複雜任務的難題，需要提升單機的自主作業能力。包括開發基於深度強化學習的決策方法、基於視覺及其他感測器多源資訊的自主環境感知與交互方法、基於神經動力學的機器人自主運動規劃方法，以及基於人工智慧的自主作業方法等，以提升智能無人系統個體的自主環境建模與定位能力、自主決策能力、自主規劃能力及自主控制能力，使智能無人系統能夠適應複雜的環境建模與定位能力、自主決策能力、自主規劃能力及自主控制能力，使智能無人系統能夠適應複雜的環境建模並開展自主作業。
⒊演算法晶片化實現高即時響應
智慧無人系統面臨的複雜環境對演算法、算力提出了較高要求，需要能即時加速運算，實現對戰場突發事件的高即時回應。為解決此問題，需要提高智慧無人系統個體演算法的晶片化水平，即開發新型架構的存算一體晶片，以提高晶片的算力與演算法晶片化水平。可研究基於人工神經技術的新型晶片，透過改變數位晶片的二進制計算方式，交換梯度訊號或權重訊號來使晶片以模擬神經元的方式進行工作，模擬大腦有效處理大數據量的並行運算流，獲得超級電腦的並行運算能力，從而極大地提升晶片的計算力與晶片化水平，解決智慧系統的高即時演算法響應。
⒋通用化的架構提升集群互通性
為提高智慧無人系統面臨高度複雜環境的適應能力，以及智慧無人系統的維修保障效率，未來智慧無人系統將繼續發展標準化的指控框架，提高人機協作的智慧性並提高系統的模組化程度。主要體現在：
①開發通用式的人工智慧框架，支援人與機器之間自主、精確、即時的良好耦合與協作關係；
②提高智慧無人系統的模組化與零件互換性，以支援在未來戰場中對智慧無人系統及其成員進行的快速維修與配置升級；
③提高資料傳輸一體化水平，以及在未來戰場上資料傳輸的抗干擾能力，降低資料的被截獲率。
⒌跨域協同打破群集應用邊界
為提高智慧無人系統在高度複雜環境下的適應能力，解決作戰時的邊界不確定難題，需要提高智慧無人系統的跨域協同作戰能力，以彌補單一作戰域能力的不足。可透過智慧無人系統的跨域協同作戰，將各個組件進行優勢互補。即利用空中無人系統的搜尋範圍大、通訊距離遠等優點，以及陸地無人系統與海洋無人系統續航時間長、穩定性強等優點，將不同組件的優勢進行組合，以增加智能無人系統的多維空間資訊感知能力，構成異質多自主體協同系統，從而提高智能無人系統完成複雜任務的能力。
⒍安全網路保障集群可靠應用
智慧無人系統在未來戰場上面臨著資訊不完整與博弈高對抗的難題，因此需要提高智慧無人系統在高對抗環境下的網路安全保障能力，提高在應對高複雜、高變化任務時的靈活性與面臨高強度網路攻擊時的穩定性。對抗環境下網路安全保障能力的提升主要體現在以下幾個方面：
①規劃合理的資料權限，以確保資料的安全性與任務執行的彈性；
②提升資訊保障能力，開發並升級智慧無人系統的資訊保障產品，備案資訊爆炸狀況的因應決策；
③增加網路的深度防禦能力，統一網路安全的標準與等級，建構網路防禦的自主性，提升網路攻擊下網路的抗打擊能力。
⒎人機混合智能提升對抗能力
為解決在未來戰場上面臨的高即時回應的難題，提高智慧無人系統在高度複雜環境下的適應能力，需要將人類與機器的優點結合，構成一種新的人機協作的混合智慧方式，即發展智慧無人系統的人機混合智慧。智慧無人系統人機混合智慧是一種由人、機、環境系統相互作用的新的物理與生物結合的智慧科學系統。針對智慧無人系統在未來戰場上所面臨的高複雜環境與高即時反應的難題，未來人機混合智慧的發展主要體現在以下幾個方面：
①資訊智能輸入。在獲取資訊的輸入端，將無人系統設備感測器客觀收集的資訊資料與作戰指揮人員的主觀感知資訊結合，構成一種多維的資訊獲取與資訊輸入方式。
②資訊智能融合。在取得多維的資料資訊後，透過將電腦的運算資料與作戰指揮人員的資訊認知融合，建構一種新的資料理解途徑。
③資訊智慧輸出。將資料資訊進行融合處理之後，將電腦的計算結果與作戰指揮人員的價值決策相互匹配，從而形成有機結合的機率化與規則化的最佳化判斷。
四、結語
智慧無人系統由於其自主性、智慧性與無人化的特點，在未來戰場上將起著日益重要的作用，智慧無人系統的發展也將帶動智慧運算、智慧交通、智慧製造、智慧醫療、類腦科學等學科領域的發展。今後，應以實際複雜環境戰場的任務需求為導向，結合人工智慧等前沿學科的先進技術，對智慧無人系統進行總體頂層規劃；在陸地、空中以及海洋無人系統中不同的無人系統作戰平台上，驗證可靠的機載智能感知與智慧運算設備，並發展可靠、穩定的無人系統自主控制、智慧感知、智慧決策與智慧互動等關鍵技術，攻克智慧無人系統的關鍵難題，不斷提升智慧無人系統的自主控制、智慧感知與智慧決策能力。
