Environment space relation reasoning method based on multi-agent debate, medium and equipment

Abstract

The invention discloses an environment space relation reasoning method based on multi-agent debate, a medium and equipment, and belongs to the technical field of large model reasoning. The environment space relation reasoning method comprises the following steps: receiving external instruction information; based on the external instruction information, controlling a plurality of agents to independently analyze, and outputting respective task planning results I; wherein the plurality of intelligent agents are composed of multi-modal large models with different system cues, and are used for representing different roles; controlling the plurality of agents to perform information interaction of respective task planning results based on an information interaction strategy, so that each of the plurality of agents outputs an updated task planning result 2 according to an information interaction result; and inputting the task planning results II output by all the agents into the judgment agent for analysis. The environment space relation reasoning method based on the multi-agent debate is high in three-dimensional space sensing capacity and high in environment space relation reasoning capacity for complex conditions, and illusion can be well resisted.

Description

Technical Field

[0001] The present application relates to the field of large model reasoning technology, and specifically to an environmental spatial relationship reasoning method, medium and device based on multi-agent debate. Background Art

[0002] An embodied robot is an intelligent entity that can interact with the environment, perceive, plan, make decisions, act, and execute tasks autonomously, just like a human. With the rapid development of large models, the feasibility and general capabilities of embodied robots have also increased significantly, and research on them has become increasingly popular.

[0003] Multimodal large models have great potential in the field of embodied robotics, but they also face some challenges and problems. First, most existing visual language models can only process two-dimensional image information, which limits their ability to perceive and understand three-dimensional space. In the real world, robots need to accurately model their surroundings in three dimensions in order to better navigate and perform tasks. However, two-dimensional image information often cannot provide enough depth information, resulting in limited ability of robots in spatial reasoning. Secondly, large models generally have hallucination problems, that is, the model may generate information or conclusions that do not match reality. This hallucination problem is particularly prominent in the decision-making process of embodied robots, because wrong information may cause the robot to take wrong actions. In addition, large models are also insufficient in reasoning ability, especially when faced with complex and uncertain environments, the model may not be able to make accurate judgments and decisions.

[0004] The existence of the above problems limits the performance of embodied robots based on multimodal large models in environmental spatial relationship reasoning. For example, in complex indoor environments, such as coffee shops, robots may find it difficult to accurately identify and understand the spatial layout, which affects their navigation and path planning capabilities. In a dynamically changing environment, the robot may make incorrect responses due to the model's hallucination problem, resulting in mission failure. To solve these problems, researchers are exploring a variety of methods. On the one hand, the robot's perception of three-dimensional space can be enhanced by integrating more advanced sensors and algorithms, such as using devices such as laser radar (LiDAR) and depth cameras to obtain richer spatial information. On the other hand, the hallucination problem can be reduced and the model's reasoning ability can be improved by improving the model architecture and training strategy. For example, by introducing technologies such as attention mechanisms and reinforcement learning, the model can be made more focused on key information and its ability to handle complex situations can be improved.

[0005] For example, Chinese patent application CN118014080A discloses a multi-embodied intelligent collaborative game decision-making method based on a large language model, including: taking embodied intelligent visual environment observation as input; parsing observable intelligent environmental features; simulating intelligent understanding and memory of the environment through a large language model based on observable environmental features; the large language model makes a comprehensive analysis based on the understanding and memory of the environment to make high-level action decisions; inter-agent communication design and management, including when to communicate, how to communicate and the content of communication, to achieve information sharing between intelligent agents; heuristic underlying action planning, mapping high-level action decisions to underlying action instruction codes; cross-platform instruction transmission, uploading the underlying action instruction codes to the collaborative game environment, and the intelligent agent takes action to complete the current moment environment deduction; this patent application uses the rich knowledge and meticulous and superior logical thinking of the large language model to achieve real-time and accurate intelligent decision-making for multi-embodied intelligent collaborative game scenarios. However, it uses the collaborative cooperation of multiple embodied intelligent agents to achieve complete visual environment detection, and there is no process of mutual verification of results between multiple intelligent agents. Therefore, the reasoning ability of the model will be relatively poor and hallucination problems are prone to occur.

[0006] Therefore, the industry urgently needs to develop a new environmental spatial relationship reasoning method based on multi-agent debate to solve the current defects and shortcomings. Summary of the invention

[0007] In view of this, the present application aims to solve at least one of the above-mentioned technical problems in the related art to a certain extent. To this end, the present application provides an environmental spatial relationship reasoning method, medium, and device based on multi-agent debate, and the provided environmental spatial relationship reasoning method based on multi-agent debate has strong three-dimensional space perception ability and strong environmental spatial relationship reasoning ability when facing complex situations.

[0008] To achieve the above objectives, in a first aspect, the present application provides an environmental spatial relationship reasoning method based on multi-agent debate, comprising the following steps:

[0009] Receive external command information;

[0010] Based on external instruction information, control multiple intelligent agents to independently analyze and output respective task planning results; wherein the multiple intelligent agents are composed of a multimodal large model with different system prompt words, which are used to represent different roles;

[0011] Controlling the plurality of intelligent agents to perform information exchange of respective task planning results among them based on the information exchange strategy, so that each of the plurality of intelligent agents outputs updated task planning result 2 according to the information exchange result;

[0012] The task planning results output by all agents are input to the referee agent for analysis, and the referee agent outputs the final task decomposition results.

[0013] Preferably, the receiving of external command information includes receiving language command information and / or two-dimensional image information currently captured by the camera.

[0014] Preferably, based on the external instruction information, the step of controlling multiple agents to independently analyze and output respective task planning results comprises:

[0015] The received language instruction information and / or the two-dimensional image information captured by the current camera are simultaneously input into all intelligent agents, so that each intelligent agent can independently perform analysis based on its own multimodal large model and output its own task planning result.

[0016] Preferably, the multiple agents include: a task planning engineer agent that is good at causal logic analysis and decomposition of complex tasks, and / or a spatial reasoning expert agent that is good at complex spatial relationship analysis and reasoning.

[0017] Preferably, in the step of controlling the multiple agents to exchange information on their respective task planning results based on an information interaction strategy, the information interaction strategy is configured to use prompt words to enable the current agent to re-analyze based on the analysis results of the remaining agents.

[0018] Preferably, the step of re-analyzing the current agent based on the analysis results of the remaining agents includes:

[0019] The current agent classifies and merges the task planning results output by other agents as new input information for the current agent and re-analyzes them.

[0020] Preferably, in the step of controlling the multiple intelligent agents to exchange information on their respective task planning results based on an information interaction strategy, the information interaction strategy is configured to use prompt words to enable the current intelligent agent to use the analysis results of the remaining intelligent agents as a reference and re-analyze.

[0021] Preferably, the multiple intelligent agents have different information interaction strategies, and the information interaction strategies of some intelligent agents are configured to use prompt words to enable the current intelligent agent to re-analyze based on the analysis results of the remaining intelligent agents; the information interaction strategies of the remaining intelligent agents are configured to use prompt words to enable the current intelligent agent to re-analyze with reference to the analysis results of the remaining intelligent agents; the different information interaction strategies have different degrees of credibility for the analysis results of the remaining intelligent agents.

[0022] In a second aspect, the present application provides a computer-readable storage medium, which stores a computer program. When the computer program runs on a computer or a processor, the computer or the processor executes any of the environmental spatial relationship reasoning methods based on multi-agent debate as described above.

[0023] In a third aspect, the present application provides a computer device comprising a memory and a processor, wherein the memory is used to store computer programs, and the processor is used to execute the environmental spatial relationship reasoning method based on multi-agent debate.

[0024] In a fourth aspect, the present application provides an intelligent device, in the control system of which the above-mentioned environmental spatial relationship reasoning method based on multi-agent debate is used to perform environmental spatial relationship reasoning.

[0025] Preferably, the intelligent device is an autonomous driving car, an autonomous driving ship or a humanoid robot, and more preferably an embodied robot.

[0026] Based on the above technical solution, it can be seen that the environmental spatial relationship reasoning method based on multi-agent debate in this application has at least one of the following beneficial effects compared with the prior art:

[0027] (1) This application uses multiple agents to independently analyze and output task plans, and then communicate and feedback with each other, which can improve the accuracy of reasoning about environmental spatial relationships. In addition, by building agents representing different roles to fully communicate, the rationality of the reasoning results is enhanced, making the robot's decision more in line with the actual situation.

[0028] (2) This application sets up an information interaction strategy to conduct information interaction on the planning results of each task. The debate and iterative reasoning process among multiple agents helps to discover and correct errors. The hallucination problem that may occur in the reasoning process of large models can be effectively alleviated through the debate mechanism of multiple agents, thereby improving the reliability of the overall reasoning process.

[0029] (3) After receiving the output of other agents, the agent of the present application can perform iterative optimization and continuously improve the quality of the reasoning results. The diverse reasoning results of different agents can inspire new thinking angles and solutions, which helps to solve complex problems.

[0030] (4) The method of the present application can flexibly adjust the roles and number of intelligent agents according to the needs of different tasks and environments, and has good flexibility and scalability.

[0031] Other features and advantages of the present application will be described in the subsequent description, and in part will become apparent from the description, or it may be understood through the implementation of the present application that the objects and other advantages of the present application can be realized and obtained through the structures particularly pointed out in the written description, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 A flowchart of the environmental spatial relationship reasoning method based on multi-agent debate for this application;

[0033] Figure 2 A schematic diagram of the decision-making process for executing the task instruction of placing blocks of different colors in a specified order for this application;

[0034] Figure 3 A schematic diagram of the information interaction strategy executed between multiple intelligent agents in this application. DETAILED DESCRIPTION

[0035] In order to make the objectives, technical solutions and advantages of the present application more clearly understood, the present application is further described in detail below in combination with specific embodiments and with reference to the accompanying drawings.

[0036] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the embodiments of the present application. The singular forms of "a", "said" and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings.

[0037] Existing embodied robots based on multimodal large models have poor environmental spatial relationship reasoning capabilities and weak three-dimensional spatial perception capabilities. The inherent hallucination problems and lack of reasoning capabilities of large models make it difficult for existing embodied robots based on multimodal large models to make correct decisions when faced with complex situations.

[0038] In order to improve the three-dimensional spatial perception ability of embodied robots based on multimodal large models and solve the problem of hallucinations that robots are prone to, the inventors conducted in-depth research on the environmental spatial relationship reasoning method of embodied robots and proposed an environmental spatial relationship reasoning method based on multi-agent debate.

[0039] By constructing intelligent agents representing different roles, allowing them to independently analyze the environment and then fully communicate with each other, and by setting information interaction strategies to exchange information on the results of their respective task planning, the debate and iterative reasoning process between multiple agents helps to discover and correct errors. The hallucination problem that may occur in the reasoning process of large models can be effectively alleviated through the debate mechanism of multiple agents, thereby improving the reliability of the overall reasoning process.

[0040] Embodiment 1

[0041] like Figure 1 As shown, this embodiment provides an environment space relationship reasoning method based on multi-agent debate, including the following steps:

[0042] S1, receiving external command information;

[0043] S2, based on external instruction information, controlling multiple agents to independently analyze and output respective task planning results; wherein the multiple agents are composed of a multimodal large model with different system prompt words, used to represent different roles;

[0044] S3, controlling the multiple agents to perform information exchange of respective task planning results among them based on the information exchange strategy, so that each of the multiple agents outputs updated task planning result 2 according to the information exchange result;

[0045] S4. Input the task planning results 2 output by all agents into the referee agent for analysis, and the referee agent outputs the final task decomposition results.

[0046] Preferably, the receiving of external command information includes receiving language command information and / or two-dimensional image information currently captured by the camera.

[0047] Preferably, the above-mentioned step of controlling multiple intelligent agents to independently perform analysis based on external command information and outputting their respective task planning results includes: simultaneously inputting the received language command information and / or the two-dimensional image information captured by the current camera into all intelligent agents, so that each intelligent agent can independently perform analysis based on its own multimodal large model and output its own task planning result.

[0048] Preferably, the multiple agents are defined as agents that can solve problems from multiple stages such as problem logic analysis, reasoning, planning, and verification, so as to alleviate the model hallucination problem.

[0049] Preferably, the multiple agents include a task planning engineer agent that is good at causal logic analysis and decomposition of complex tasks, and / or a spatial reasoning expert agent that is good at complex spatial relationship analysis and reasoning, and other agents with multiple virtual "personalities". Specifically, engineers focus on thinking about the feasibility and effectiveness of planning, while scientists or experts focus on the reliability of the logical relationship of planning.

[0050] This application utilizes a role-playing method to enable different intelligent agents to focus on different input information and use different reasoning logic to produce multi-angle and diversified reasoning results, which can effectively alleviate the "hallucination" of large models and generate more accurate task decomposition results.

[0051] Preferably, in the step of controlling the information interaction between multiple agents based on the information interaction strategy for the respective task planning results, the information interaction strategy is configured to use prompt words to make the current agent re-analyze based on the analysis results of the remaining agents. Further, the step of the current agent re-analyzing based on the analysis results of the remaining agents includes: the current agent classifies and merges the task planning results output by the remaining agents and uses them as new input information for the current agent, and re-analyzes them. Among them, classifying and merging the task planning results output by the remaining agents can be to classify and merge the same or similar task planning results, and to perform corresponding weight distribution according to the number of the same or similar task planning results and / or their credibility. If there are conflicts or inconsistencies in the outputs of different agents, the current agent needs to have a mechanism to resolve these conflicts, such as through voting, priority rules or further logical reasoning.

[0052] Furthermore, the task planning results 1 output by the remaining intelligent agents can be classified and merged according to the task decomposition steps in the task planning results 1 output by the remaining different intelligent agents, and the same or similar task decomposition steps can be classified and merged, and corresponding weights can be allocated according to the number of the same or similar task decomposition steps and / or their credibility.

[0053] Specific as Figure 2 As shown, a schematic diagram of the decision-making process of the present application for executing a task instruction to place blocks of different colors in a specified order is shown. The task instruction issued is: place the blocks in the picture in green, yellow, blue and red in order from bottom to top. The task instruction is issued to three different agents. In the first round, each agent analyzes independently and outputs its own task planning result one, but the results output by each agent in the first round may be different. After the first round, multiple agents interact with each other based on the information interaction strategy. All agents receive the reasoning results from the remaining agents and update their second round answers. At this time, the output results of the second round are not necessarily consistent, but each agent tends to make more accurate reasoning after careful consideration, and the results may be more similar. After the second round, the task planning result two output by all agents will be input into the referee agent for analysis, and the referee agent will output the final task disassembly result.

[0054] like Figure 3As shown, a schematic diagram of executing information interaction strategies between multiple agents of the present application is shown. In the step of controlling the information interaction between multiple agents based on the information interaction strategy for the respective task planning results, the information interaction strategy is configured to use prompt words to make the current agent use the analysis results of the remaining agents as a reference and re-analyze. It can be further selected that the multiple agents can also have different information interaction strategies, and the information interaction strategies of some agents are configured to use prompt words to make the current agent re-analyze based on the analysis results of the remaining agents; the information interaction strategies of the remaining agents are configured to use prompt words to make the current agent use the analysis results of the remaining agents as a reference and re-analyze; the different information interaction strategies have different credibility for the analysis results of the remaining agents. By controlling the "credibility" of each agent's answer to the other agents during the debate, the intensity of the debate is controlled, which will have different effects on the final debate results. After receiving the output of other agents, the agent of the present application can perform iterative optimization and continuously improve the quality of the reasoning results. The diversified reasoning results of different agents can inspire new thinking angles and solutions, help solve complex problems, and further enhance the reasoning ability of the embodied robot.

[0055] Embodiment 2

[0056] Based on the same technical concept, the embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program runs on a computer or a processor, the computer or the processor executes the steps of the above-mentioned method for reasoning the spatial relationship of the embodied robot environment based on multi-agent debate. And realize the following functions: by constructing multiple agents representing different roles to perform independent analysis and output task planning, and then communicate and feedback with each other based on information interaction strategies, the accuracy of reasoning about the spatial relationship of the environment can be improved, and the possible illusion problem can be alleviated. And by constructing agents representing different roles to fully communicate, the rationality of the reasoning results is enhanced, so that the robot's decision is more in line with the actual situation. This application gives full play to the advantages of multi-agent game technology in reasoning scenarios, and uses multiple agents playing different roles to fully communicate to enhance the rationality and reliability of the reasoning results; at the same time, the interaction of virtual agents given multiple roles can improve the lack of perception ability of multimodal large models for three-dimensional space. This solution has obvious advantages in reasoning about spatial relationships in the embodied robot environment.

[0057] Embodiment 3

[0058] Based on the same technical concept, the embodiment of the present application also provides a computer device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to implement the steps of the above-mentioned embodied robot environment space relationship reasoning method based on multi-agent debate. And realize the following functions: by constructing multiple agents representing different roles to perform independent analysis and output task planning, and then communicate and feedback with each other based on information interaction strategies, the reasoning accuracy of the environment space relationship can be improved, and the possible hallucination problem can be alleviated. And by constructing agents representing different roles to fully communicate, the rationality of the reasoning results is enhanced, making the robot's decision more in line with the actual situation.

[0059] Embodiment 4

[0060] Based on the same technical concept, the embodiment of the present application also provides an embodied robot based on multi-agent debate, which can realize the above-mentioned embodied robot environment space relationship reasoning method based on multi-agent debate. In this embodiment, firstly, a plurality of virtual "personalities" of agents are constructed using large model prompt words, and each agent is composed of a multimodal large model with different system prompt words. For example, it can be clearly introduced in the system prompt words: "You are a task planning engineer, good at causal logic analysis and disassembly of complex tasks", or "You are a spatial reasoning expert, good at complex spatial relationship analysis and reasoning", so that different agents can pay attention to different input content in the reasoning process, adopt different reasoning logic to output diversified reasoning results, which is crucial for further communication and feedback based on the reasoning results to continuously correct errors. When the embodied robot receives the language instruction, the two-dimensional image and language instruction taken by the current camera are simultaneously input into multiple agents, and all agents independently make their own analysis and output task planning; then, each agent receives the output of the remaining agents as external feedback, and uses prompt words to guide the re-reasoning. For example, an instruction can be added at the end of the agent input information: "Use the above results from other agents as a reference, rethink your answer and give an updated reasoning result." Finally, a "referee" agent is constructed to analyze and judge the debate results, and all the outputs of the last round of all agents participating in the debate are combined as the input of the "referee" agent. At the same time, an instruction is added: "Analyze the above reasoning results from all agents, combine visual information, and give the final task decomposition result"; finally, the output of the "referee" agent is used as the task decomposition result. The embodied robot based on multi-agent debate provided in this embodiment can perform iterative optimization after receiving the output of other agents, and continuously improve the quality of the reasoning results. The diversified reasoning results of different agents can inspire new thinking angles and solutions, and can solve complex problems.

[0061] The above describes specific embodiments of the present application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0062] The method steps in the embodiments of the present application can be implemented in hardware or in software instructions that can be executed by a processor. The software instructions can be composed of corresponding software modules, and the software modules can be stored in random access memory, flash memory, read-only memory, programmable read-only memory, erasable programmable read-only memory, electrically erasable programmable read-only memory, registers, hard disks, mobile hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor so that the processor can read information from the storage medium and write information to the storage medium. The storage medium can also be a component of the processor. The processor and the storage medium can be located in an ASIC. In addition, the ASIC can be located in a base station, a core network device, or a network element management system. The processor and the storage medium can also exist as discrete components in a base station, a core network device, or a network element management system.

[0063] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented by software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer programs or instructions. When the computer program or instruction is loaded and executed on a computer, the process or function described in the embodiment of the present application is executed in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, a network device, a user device or other programmable device. The computer program or instruction may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer program or instruction may be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired or wireless means. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server, data center, etc. that integrates one or more available media. The available medium may be a magnetic medium, for example, a floppy disk, a hard disk, a tape; it may also be an optical medium, for example, a digital video disc; it may also be a semiconductor medium, for example, a solid-state hard disk. The computer-readable storage medium may be a volatile or nonvolatile storage medium, or may include both volatile and nonvolatile types of storage media.

[0064] The above description is only a preferred embodiment of the embodiments of the present application and is not intended to limit the embodiments of the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present application should be included in the scope of protection of the embodiments of the present application.

Claims

1. A method for reasoning about environmental spatial relations based on multi-agent debate, characterized in that: include: Receive external command information; Based on external instruction information, control multiple intelligent agents to independently analyze and output respective task planning results; wherein the multiple intelligent agents are composed of a multimodal large model with different system prompt words, which are used to represent different roles; Controlling the plurality of intelligent agents to perform information exchange of respective task planning results among them based on the information exchange strategy, so that each of the plurality of intelligent agents outputs updated task planning result 2 according to the information exchange result; The task planning results output by all agents are input to the referee agent for analysis, and the referee agent outputs the final task decomposition results.

2. The environmental spatial relationship reasoning method according to claim 1, characterized in that: The receiving of external instruction information includes receiving language instruction information and / or two-dimensional image information currently captured by the camera.

3. The environmental spatial relationship reasoning method according to claim 2, characterized in that: Based on external command information, the steps of controlling multiple agents to independently analyze and output their respective task planning results include: The received language instruction information and / or the two-dimensional image information captured by the current camera are simultaneously input into all intelligent agents, so that each intelligent agent can independently perform analysis based on its own multimodal large model and output its own task planning result.

4. The environmental spatial relationship reasoning method according to claim 1, characterized in that: The multiple agents include: a task planning engineer agent that is good at causal logic analysis and decomposition of complex tasks, and / or a spatial reasoning expert agent that is good at complex spatial relationship analysis and reasoning.

5. The environmental spatial relationship reasoning method according to claim 1, characterized in that: In the step of controlling the multiple agents to exchange information on their respective task planning results based on an information interaction strategy, the information interaction strategy is configured to use prompt words to enable the current agent to re-analyze based on the analysis results of the remaining agents.

6. The environmental spatial relationship reasoning method according to claim 5, characterized in that: The steps for the current agent to re-analyze the results of the analysis of other agents include: The current agent classifies and merges the task planning results output by other agents as new input information for the current agent and re-analyzes them.

7. The environmental spatial relationship reasoning method according to claim 1, characterized in that: In the step of controlling the multiple agents to exchange information on their respective task planning results based on an information interaction strategy, the information interaction strategy is configured to use prompt words to enable the current agent to use the analysis results of the remaining agents as a reference and re-analyze.

8. The environmental spatial relationship reasoning method according to claim 1, characterized in that: The multiple intelligent agents have different information interaction strategies. The information interaction strategies of some intelligent agents are configured to use prompt words to enable the current intelligent agent to re-analyze based on the analysis results of the remaining intelligent agents; the information interaction strategies of the remaining intelligent agents are configured to use prompt words to enable the current intelligent agent to re-analyze with reference to the analysis results of the remaining intelligent agents; the different information interaction strategies have different degrees of credibility for the analysis results of the remaining intelligent agents.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when run on a computer or a processor, enables the computer or the processor to execute the environmental spatial relationship reasoning method based on multi-agent debate as described in any one of claims 1 to 8.

10. A computer device comprising a memory and a processor, wherein the memory is used to store a computer program, wherein: The processor is used to execute the computer program to implement the environmental spatial relationship reasoning method based on multi-agent debate as described in any one of claims 1 to 8.

11. A smart device, characterized in that: The control system of the intelligent device adopts the environmental spatial relationship reasoning method based on multi-agent debate as described in any one of claims 1 to 8 to perform environmental spatial relationship reasoning.

12. The smart device according to claim 11, characterized in that: The intelligent device is an autonomous driving car, an autonomous driving ship or a humanoid robot.

Images