Method and device for improving reasoning ability of navigation robot based on thought chain

Abstract

The present invention discloses a method and device for improving the reasoning ability of a navigation robot based on a thought chain. The present invention reads the graph structure of the navigation scene, extracts the current observation information, forms a multimodal prompt word in combination with the historical path, prompts the multimodal large model to generate multiple reference paths, and then uses the path selection large model to select the optimal path. Furthermore, for the case where the model selection options are different from the output, the option check large model is used to check the options of the optimal path. Specifically, by loading the navigation graph structure data, an action description is generated in combination with the candidate viewpoints; multiple paths are generated by the beam search method, and then the path selection large model is prompted to evaluate the path and output the optimal decision. For the case where the model selection options are different from the output, the option check large model is used to correct the options of the optimal path. The present invention significantly reduces the loss of navigation information and improves the multimodal reasoning ability, path planning accuracy and generalization performance of the model.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence application and multimodal information fusion, and in particular relates to a method and device for improving the reasoning ability of a navigation robot based on thought chain. Background Art

[0002] Artificial intelligence (AI) is rapidly developing, and the design of intelligent agents is moving towards multi-tasking and multimodal generalization. Within this trend, embodied intelligence (Embodied Intelligence), a new intelligent paradigm that integrates perception, thinking, and action, has become a research hotspot. This approach focuses specifically on agents' ability to interact with their environment, completing complex tasks through dynamic perception and decision-making. This is particularly relevant in application scenarios such as smart homes, service robots, and autonomous driving. The core concept of this technology is to endow agents with human-like perception and action capabilities, enabling them to autonomously complete tasks in complex, open environments, overcoming the limitations of traditional AI in dynamic environments.

[0003] At the same time, the rise of large multimodal models provides strong technical support for embodied intelligence. By integrating perceptual modalities such as vision, language, and sound, these models demonstrate remarkable cross-modal reasoning and task execution capabilities. For example, Vision-and-Language Navigation (VLN), a key research area in embodied intelligence, requires agents to plan and navigate paths in a visual environment under the guidance of language instructions, placing significant demands on the model's cross-modal reasoning and environmental adaptability.

[0004] Before the advent of the large model era, VLN research focused on aligning information from different modalities through representation learning and optimizing action selection through policy learning. These studies leveraged pre-trained vision and language models (such as ResNet and BERT) combined with techniques such as cross-modal alignment and imitation learning to address the performance gap between partially known and unknown environments.

[0005] With the introduction of large language models (LLMs) and large multimodal models (MLLMs), VLN research has entered a new phase. Through multimodal pre-training, these models significantly improve their ability to understand and generalize complex environments. For example, systems based on large models can translate visual input into textual descriptions and combine them with instructions to predict actions, significantly improving the accuracy of task planning and execution.

[0006] To further enhance reasoning capabilities, researchers have proposed a CoT-based approach that incrementally completes complex tasks through explicit reasoning chains. As this approach expands from single-modality to multimodality (such as combining vision and language), it enables logical deduction and efficient execution of complex tasks. For example, models such as EmbodiedGPT introduce an embodied CoT reasoning model, empowering agents with dynamic planning and real-time decision-making capabilities, opening up new directions in multimodal embodied intelligence.

[0007] Despite significant technological advances, current research still faces the following key challenges: 1. Deep fusion of multimodal information: How to efficiently integrate information from visual, language, and other modalities to achieve real-time environmental perception and dynamic planning. 2. Generalization: How to enhance the agent's performance in unseen environments given data scarcity and environmental diversity.

[0008] Therefore, the existing thought chain method has failed to fully unleash the potential of large multimodal models in terms of final effect, and there is still room for improvement in reasoning effect. Summary of the Invention

[0009] The purpose of the present invention is to address the deficiencies of the existing technology and provide a method and device for improving the reasoning ability of a navigation robot based on thought chain.

[0010] The object of the present invention is achieved through the following technical solution: a method for improving the reasoning ability of a navigation robot based on thought chain, comprising the following steps:

[0011] (1) Load the navigation graph of the current scene and obtain the graph structure; obtain the observation information of the current scene and generate an action prompt list; and generate trajectory text and graph structure text based on the graph structure;

[0012] (2) Combine the instructions, historical paths, the previous plan, the action prompt list, the trajectory text, and the graph structure text into the first prompt word;

[0013] (3) The first prompt word and the current environment view are then input into the multimodal large model, which then outputs multiple thoughts and the next step plan corresponding to each thought;

[0014] (4) The obtained multiple thoughts, the next plan corresponding to each thought, and the first prompt word are spliced ​​into the second prompt word and input into the optimal path selection model, and then the optimal thought and the next plan corresponding to the optimal thought are output;

[0015] (5) The optimal thinking, the next step plan corresponding to the optimal thinking, and the obtained action prompt list are spliced ​​into the third prompt word and input into the option check model, and then the action prompt corresponding to the optimal thinking and the motion description options are output; the robot moves to the next location according to the motion description options;

[0016] (6) If the target location is not reached, repeat steps (1) to (5) until the robot reaches the target location.

[0017] Furthermore, the method of loading the navigation graph of the current scene and obtaining the graph structure is as follows: loading the navigation graph of the current scene, obtaining the connectivity data of each scene in JSON format and reading it using the Dijkstra algorithm, screening out all passable perspectives, calculating the shortest path and corresponding distance of each pair of perspectives and storing them in the graph structure, treating each perspective as a node in the graph structure, treating the connection relationship between the two perspectives as an edge in the graph structure, and treating the distance between the two perspectives as the weight of the corresponding edge.

[0018] Furthermore, the obtaining of observation information of the current scene is specifically as follows: obtaining the instruction ID, scan ID and position information of the current view of the current scene; then obtaining all candidate viewpoints of the current view through the graph structure, continuously traversing each candidate viewpoint corresponding to the current viewpoint, generating information of each candidate viewpoint and storing it in a candidate viewpoint list, wherein the information of the candidate viewpoint includes feature information and position information of the candidate viewpoint; and taking the observation information including the instruction ID, scan ID, position information of the current viewpoint and the candidate viewpoint list as the observation information of the current scene.

[0019] Furthermore, based on the observation information, an action prompt list is generated, specifically: based on the candidate perspective list in the observation information, the horizontal perspective and rotational perspective of the current perspective are compared with the horizontal perspective and pitch angle of each candidate perspective to obtain action prompts between the current perspective and each candidate perspective, and each action prompt is labeled with a letter to generate an action prompt list; the action prompts include turning left, going straight, turning right, going upstairs and going downstairs.

[0020] Furthermore, the generation of trajectory text and graph structure text according to the graph structure is specifically as follows: first, the trajectory text is constructed according to the shortest path of each pair of perspectives in the graph structure; then, the graph structure text is spliced ​​out through the connection relationship between each perspective and other perspectives in the graph structure.

[0021] Furthermore, the multimodal large model is Qwen2-VL-7B; the optimal path selection large model is Qwen2.5-0.5B; and the option check large model is Qwen2.5-0.5B.

[0022] The present invention also includes a device for improving the reasoning ability of a navigation robot through multi-step planning based on a multimodal thinking chain, comprising a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, it is used for the above-mentioned method for improving the reasoning ability of a navigation robot through multi-step planning based on a multimodal thinking chain.

[0023] The present invention also includes a computer-readable storage medium having a program stored thereon. When the program is executed by a processor, the method of improving the reasoning ability of a navigation robot by multi-step planning based on a multimodal thinking chain is implemented.

[0024] The beneficial effects of the present invention are:

[0025] 1) This paper significantly improves the reasoning ability of the multimodal large model (MLLM) in vision-language navigation (VLN) tasks by combining multimodal thought chains and multi-step planning mechanisms;

[0026] 2) By introducing a multimodal thinking chain, this invention reduces information loss in multimodal conversion by directly processing environmental images rather than relying solely on language descriptions during the input process, enabling the model to more accurately understand the agent's environment and improving the reliability and robustness of task execution;

[0027] 3) This invention uses a multi-step planning mechanism as its core, viewing the model's diverse thinking as an opportunity to improve its reasoning capabilities. By reflecting on the self-generated reasoning paths, the model can significantly consider the reasons behind each path. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 A flowchart of a method for improving the reasoning ability of a navigation robot based on thought chain;

[0029] Figure 2 This is a schematic diagram of constructing the first prompt word in Example 1;

[0030] Figure 3 Schematic diagram of the multimodal large model, the optimal path selection large model, and the option checking large model in Example 1;

[0031] Figure 4 This is a structural diagram of a device for improving the reasoning ability of a navigation robot based on thought chain. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to illustrate the present invention, rather than to represent all embodiments. All other embodiments obtained by persons of ordinary skill in the art based on the embodiments of the present invention without creative work are within the scope of protection of the present invention.

[0033] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. As used in this application and the appended claims, the singular forms "a," "an," "the," and "the" are intended to include the plural forms, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0034] The present invention will be described in detail below with reference to the accompanying drawings. Unless there is any conflict, the features of the following embodiments and implementations may be combined with each other.

[0035] The present invention constructs a multi-step reasoning process based on a multimodal thinking chain and a multi-step planning mechanism, utilizing the graph structure, historical path, and environmental observation information in the navigation scene, while integrating multimodal inputs such as vision and language to achieve a deep understanding and dynamic planning of the robot's navigation task. In the method of the present invention, the scene data is first preprocessed to obtain the navigation map and candidate perspective information, and multimodal analysis is performed in combination with environmental images, instruction texts, etc., and then action prompts and path decisions are continuously generated in the thinking chain reasoning process; through the multi-step planning mechanism, multiple thoughts can be output simultaneously and the optimal path planning can be screened out, thereby effectively filtering and identifying redundant or uncertain information in the multimodal data. Since multimodal features such as environmental images and language instructions are fully combined in each step, the present invention can explore the complementarity and correlation between modalities while reducing information loss, greatly improving the reasoning accuracy and generalization ability of the navigation robot, and providing a more flexible and efficient solution for the application of multimodal large models in embodied intelligence scenarios.

[0036] Example 1: Figure 1 、 Figure 2 and Figure 3 As shown, the present invention provides a method for improving the reasoning ability of a navigation robot based on thought chain, comprising the following steps:

[0037] (1) Load the navigation graph of the current scene and obtain the graph structure; obtain the observation information of the current scene and generate an action prompt list; and generate trajectory text and graph structure text based on the graph structure.

[0038] The method of loading the navigation graph of the current scene and obtaining the graph structure is as follows: loading the navigation graph of the current scene, obtaining the connectivity data of each scene in JSON format and reading it using the Dijkstra algorithm, screening out all accessible perspectives, calculating the shortest path and corresponding distance for each pair of perspectives and storing them in the graph structure, treating each perspective as a node in the graph structure, treating the connection between the two perspectives as an edge in the graph structure, and treating the distance between the two perspectives as the corresponding edge weight. In this way, the shortest path planning information between each perspective node in each scene is obtained, providing structured data support for subsequent multimodal navigation and decision-making.

[0039] The obtaining of the observation information of the current scene is specifically as follows: obtaining the instruction ID (instr_id), scan ID (scan) and position information (position, heading, elevation) of the current scene; then obtaining all candidate viewpoints of the current viewpoint through the graph structure, continuously traversing each candidate viewpoint corresponding to the current viewpoint, generating information of each candidate viewpoint and storing it in a candidate viewpoint list (candidate), wherein the information of the candidate viewpoint includes feature information and position information of the candidate viewpoint; and taking the observation information including the instruction ID, scan ID, position information of the current viewpoint and the candidate viewpoint list as the observation information of the current scene.

[0040] The method of generating an action prompt list based on the observation information is as follows: based on the candidate perspective list in the observation information, the horizontal perspective and rotational perspective of the current perspective are compared with the horizontal perspective and pitch angle of each candidate perspective to obtain action prompts between the current perspective and each candidate perspective, and each action prompt is labeled with a letter to generate an action prompt list; the action prompts include turning left, going straight, turning right, going upstairs and going downstairs.

[0041] For example, when the difference between the horizontal angle of view of a candidate perspective and the horizontal angle of view of the current perspective is positive, the action prompt for the candidate perspective can be determined to be a right turn; when the difference between the horizontal angle of view of the candidate perspective and the horizontal angle of view of the current perspective is negative, the action prompt for the candidate perspective can be determined to be a left turn; when the difference between the horizontal angle of view of the candidate perspective and the horizontal angle of view of the current perspective is 0, the action prompt for the candidate perspective can be determined to be walking straight; when the rotation angle of the candidate perspective is greater than 0, the action prompt for the candidate perspective can be determined to be going upstairs; when the rotation angle of the candidate perspective is less than 0, the action prompt for the candidate perspective can be determined to be going downstairs. Based on the direction descriptions determined by these judgments, when generating action prompts, it is only necessary to combine these direction descriptions (such as "turn left" or "go upstairs") with the relevant information of the target node to clearly and unambiguously express the actions required to move from the current perspective to the next target perspective.

[0042] The generation of trajectory text and graph structure text according to the graph structure is specifically as follows: first, the trajectory text is constructed according to the shortest path of each pair of perspectives in the graph structure; then, the graph structure text is spliced ​​out through the connection relationship between each perspective and other perspectives in the graph structure.

[0043] (2) Combine the instructions, historical paths, the previous plan, the action prompt list, the trajectory text, and the graph structure text into the first prompt word, specifically:

[0044] First, a list of action options with alphabetical labels (e.g., "A. turn left to Place 2") is generated for each candidate perspective, with a "stop" option inserted at the beginning where applicable. Next, the trajectory text and graph structure text for each example in the current batch are obtained. This information describes the nodes visited during the navigation task, the connections between nodes, and unexplored ghost nodes. Then, depending on whether it is the first step (t == 0), an empty or existing history record is appended to the first prompt. The instructions, historical paths, the previous planned plan, and the action prompt list are also integrated into the first prompt. This is supplemented with a task description that includes background settings, map instructions, execution requirements, and output format. This ensures that the returned prompt contains a complete and coherent navigation context, providing the necessary environmental information and decision-making clues for subsequent multimodal thought chain reasoning.

[0045] (3) The first prompt word and the current environment view are then input into the multimodal large model, and then multiple thoughts and the next plan corresponding to each thought are output.

[0046] (4) The obtained multiple thoughts, the next plan corresponding to each thought, and the first prompt word are spliced ​​into the second prompt word and input into the optimal path selection model, and then the optimal thought and the next plan corresponding to the optimal thought are output;

[0047] (5) The optimal thinking, the next step plan corresponding to the optimal thinking, and the obtained action prompt list are spliced ​​into the third prompt word and input into the option check model, and then the action prompt corresponding to the optimal thinking and the motion description option are output; the robot moves to the next location according to the motion description option.

[0048] The multimodal large model is Qwen2-VL-7B; the optimal path selection large model is Qwen2.5-0.5B; and the option check large model is Qwen2.5-0.5B.

[0049] (6) If the target location is not reached, repeat steps (1) to (5) until the robot reaches the target location.

[0050] Example 2: This example relates to a device for improving the reasoning ability of a navigation robot based on a thought chain, comprising a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, it is used for a method for improving the reasoning ability of a navigation robot based on a thought chain in the above-mentioned example 1; the device embodiment can be applied to any device with data processing capabilities, and the any device with data processing capabilities can be a device or apparatus such as a computer.

[0051] like Figure 4 At the hardware level, the knowledge distillation device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory. Of course, it may also include other hardware required for the business. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to achieve the above Figure 1 Of course, in addition to software implementation, the present invention does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc., that is, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0052] Improvements to a technology can be clearly categorized as either hardware improvements (for example, improvements to circuit structures like diodes, transistors, and switches) or software improvements (improvements to process flows). However, with technological advancements, many process flow improvements can now be considered direct improvements to hardware circuit structures. Designers almost always create the corresponding hardware circuit structure by programming the improved process flow into the hardware circuit. Therefore, it cannot be said that a process flow improvement cannot be implemented using physical hardware modules. For example, a programmable logic device (PLD), such as a field programmable gate array (FPGA), is an integrated circuit whose logical function is determined by user programming. Designers can integrate a digital system onto a PLD through their own programming, eliminating the need for a chip manufacturer to design and manufacture a dedicated integrated circuit chip. Moreover, nowadays, instead of manually fabricating integrated circuit chips, this programming is mostly performed using "logic compiler" software. This is similar to the software compiler used during program development. Before compilation, the original code must also be written in a specific programming language, called a hardware description language (HDL). There is not just one HDL, but many, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art will also understand that by simply programming the method flow in one of these hardware description languages ​​and then programming it into an integrated circuit, a hardware circuit that implements the logic method flow can be easily obtained.

[0053] The controller can be implemented in any suitable manner. For example, the controller can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, an application-specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the memory control logic. Those skilled in the art will also appreciate that, in addition to implementing the controller purely in computer-readable program code, the controller can also be implemented in the form of logic gates, switches, an application-specific integrated circuit, a programmable logic controller, an embedded microcontroller, etc. by logically programming the method steps. Therefore, such a controller can be considered a hardware component, and the means for implementing the various functions included therein can also be considered as structures within the hardware component. Alternatively, the means for implementing the various functions can be considered both a software module implementing the method and a structure within the hardware component.

[0054] The systems, devices, modules, or units described in the above embodiments may be implemented by computer chips or entities, or by products having certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0055] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0056] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code.

[0057] The present invention may be described in the general context of computer-executable instructions, such as program modules, executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, and the like that perform specific tasks or implement specific abstract data types. The present invention may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media, including storage devices.

[0058] Example 3: An embodiment of the present invention further provides a computer-readable storage medium having a program stored thereon. When the program is executed by a processor, the method of improving the reasoning ability of a navigation robot based on a thought chain in the above-mentioned Example 1 is implemented.

[0059] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for improving the reasoning ability of a navigation robot based on thought chain, characterized in that: The following steps are involved: (1) Load the navigation graph of the current scene and obtain the graph structure; obtain the observation information of the current scene and generate an action prompt list; And generate trajectory text and graph structure text according to the graph structure; (2) Combine the instructions, historical paths, the previous plan, the action prompt list, the trajectory text, and the graph structure text into the first prompt word; (3) The first prompt word and the current environment view are then input into the multimodal large model, which then outputs multiple thoughts and the next step plan corresponding to each thought; (4) The obtained multiple thoughts, the next plan corresponding to each thought, and the first prompt word are spliced ​​into the second prompt word and input into the optimal path selection model, and then the optimal thought and the next plan corresponding to the optimal thought are output; (5) The optimal thinking, the next step plan corresponding to the optimal thinking, and the obtained action prompt list are spliced ​​into the third prompt word and input into the option check model, and then the action prompt corresponding to the optimal thinking and the motion description options are output; the robot moves to the next location according to the motion description options; (6) If the target location is not reached, repeat steps (1) to (5) until the robot reaches the target location.

2. The method for improving the reasoning ability of a navigation robot based on thought chain according to claim 1, characterized in that: The method of loading the navigation graph of the current scene and obtaining the graph structure is as follows: loading the navigation graph of the current scene, obtaining the connectivity data of each scene in JSON format and reading it using the Dijkstra algorithm, screening out all passable perspectives, calculating the shortest path and corresponding distance of each pair of perspectives and storing them in the graph structure, treating each perspective as a node in the graph structure, treating the connection relationship between the two perspectives as an edge in the graph structure, and treating the distance between the two perspectives as the weight of the corresponding edge.

3. The method for improving the reasoning ability of a navigation robot based on thought chain according to claim 2 is characterized in that: The obtaining of observation information of the current scene is specifically as follows: obtaining the instruction ID, scan ID and position information of the current view of the current scene; then obtaining all candidate viewpoints of the current view through the graph structure, continuously traversing each candidate viewpoint corresponding to the current viewpoint, generating information of each candidate viewpoint and storing it in a candidate viewpoint list, wherein the information of the candidate viewpoint includes feature information and position information of the candidate viewpoint; and using the observation information including the instruction ID, scan ID, position information of the current viewpoint and the candidate viewpoint list as the observation information of the current scene.

4. The method for improving the reasoning ability of a navigation robot based on thought chain according to claim 3 is characterized in that: The generating of the action prompt list is specifically as follows: based on the candidate perspective list in the observation information, the horizontal perspective and rotational perspective of the current perspective are compared with the horizontal perspective and pitch angle of each candidate perspective to obtain the action prompts between the current perspective and each candidate perspective, and each action prompt is labeled with a letter to generate an action prompt list; the action prompts include turning left, going straight, turning right, going upstairs and going downstairs.

5. The method for improving the reasoning ability of a navigation robot based on thought chain according to claim 4 is characterized in that: The generation of trajectory text and graph structure text according to the graph structure is specifically as follows: first, the trajectory text is constructed according to the shortest path of each pair of perspectives in the graph structure; then, the graph structure text is spliced ​​out through the connection relationship between each perspective and other perspectives in the graph structure.

6. The method for improving the reasoning ability of a navigation robot based on thought chain according to claim 5 is characterized in that: The multimodal large model is Qwen2-VL-7B; the optimal path selection large model is Qwen2.5-0.5B; and the option check large model is Qwen2.5-0.5B.

7. A device for improving the reasoning ability of a navigation robot based on thought chain, characterized in that: It includes a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, it is used to implement a method for improving the reasoning ability of a navigation robot based on a thought chain as described in any one of claims 1 to 6.

8. A computer-readable storage medium, characterized in that A program is stored thereon, and when the program is executed by the processor, the method for improving the reasoning ability of a navigation robot based on a thought chain as described in any one of claims 1 to 6 is implemented.

Images