Robot autonomous navigation system based on AI large model

Abstract

The invention provides a robot autonomous navigation system based on an AI large model, and relates to the technical field of artificial intelligence AI large models, and the system comprises a man-machine interaction module, a body perception module, a body thinking module, a body execution module, and an interaction unified power supply system. An environment is interactively sensed through a body sensing module, namely a visual (laser radar) sensor, a touch (pressure) sensor and the like of a robot, and a body sensing model is constructed. The AI large model with the thinking module is responsible for processing perception information and performing reasoning and decision making. And the body execution module (robot execution mechanism) is responsible for executing corresponding actions according to decisions, so that the environment perception and task decision capabilities of the robot are effectively improved, and the autonomous navigation capability of the robot in a complex environment is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence (AI) large models, and particularly to a robot autonomous navigation system based on an AI large model. Background Art

[0002] The combination of AI large models and robots is not only a current hotspot but also an important direction for future intelligent revolutions. Embodied intelligence is the product of the combination of AI large models and robot entities. Embodied intelligence is an intelligent system that realizes environmental perception, information cognition, autonomous decision-making, and action through the interaction between the robot and the environment, and can achieve intelligent growth and action adaptation from experience feedback.

[0003] The AI large model provides powerful cognitive capabilities for embodied intelligence, uses the robot's perception system to obtain surrounding environment information, and plans the optimal navigation path through the reasoning and decision-making assistance of the large model. Such robots have stronger interactivity and adaptability and can play a better role in the complex and ever-changing real world. Summary of the Invention

[0004] Aiming at the deficiencies of the prior art, the present invention provides a robot autonomous navigation system based on an AI large model, which combines the LLaMA large model and its fine-tuning technology with the robot autonomous navigation system. Through embodied perception, that is, the robot's vision and other sensors interact to perceive the environment, an embodied perception module is constructed; the AI large model is responsible for processing the perception information and performing reasoning and decision-making, and an embodied thinking module is constructed; while the embodied execution module (robot actuator) is responsible for performing corresponding actions according to the decision, thereby realizing more intelligent robot autonomous navigation.

[0005] A robot autonomous navigation system based on an AI large model includes a human-computer interaction module, an embodied perception module, an embodied thinking module, an embodied execution module, and an interactive unified power supply system;

[0006] The human-computer interaction module includes a display screen and a voice broadcast module. The embodied perception module includes a 2D lidar and a CAN to USB module. The embodied thinking module is an industrial computer. The embodied execution module includes Mecanum wheels, a direct current brushless motor, and an Arduino brushless motor controller. The interactive unified power supply system includes a 24V 4000mAh lithium battery, a 24V to 19V voltage regulation module, a 24V to 12V voltage regulation module, a 24V to 5V voltage regulation module, and an emergency stop switch;

[0007] The embodied perception module obtains the surrounding environment information through a 2D lidar. The industrial control computer performs coordinate transformation on the data obtained by the 2D lidar, converting the obstacle distance information in the lidar coordinate system into the obstacle distance information in the robot coordinate system. A local LLaMA large model arranged under the ubuntu system is built into the industrial control computer, and the obstacle distance information is provided to the local LLaMA large model to achieve end-to-end autonomous navigation planning of the robot;

[0008] In the embodied execution module, Mecanum wheels are used to achieve the omnidirectional movement of the robot, and a direct current brushless motor is used to provide power for the chassis; The Arduino brushless motor controller performs FOC vector control on the direct current brushless motor. Arduino communicates with the industrial control computer through a CAN to USB module and receives the speed control signal from the industrial control computer to control the movement of the chassis.

[0009] The interactive unified power supply system is used to supply power to the entire system. A 24V 4000mah lithium battery is used to supply power to the entire system, directly supplying power to the direct current brushless motor. The 24V 4000mah lithium battery supplies power to the industrial control computer through a 24V to 19V voltage stabilizing module, and the industrial control computer is responsible for supplying power to the display screen and the voice broadcast module; It supplies power to the 2D lidar through a 24V to 12V voltage stabilizing module; It supplies power to the Arduino brushless motor controller through a 24V to 5V voltage stabilizing module. An emergency stop switch is used to start and stop the power supply of the entire system;

[0010] The 2D lidar communicates with the industrial control computer through the USB2.0 protocol. The industrial control computer performs coordinate transformation according to the set relative position relationship between the 2D lidar and the chassis of the Mecanum wheels, converts the sensor data of the 2D lidar into the sensor data of the robot itself, and provides it to the local LLaMA large model for interaction between the large model and the external environment.

[0011] The local LLaMA large model controls the posture of the robot according to the obstacle distance information transmitted back by the 2D lidar. The industrial control computer sends the speed command to the Arduino brushless motor controller through the CAN to USB module. The Arduino brushless motor controller sends the posture data of the chassis and the feedback data of the direct current brushless motor to the industrial control computer, and the local LLaMA large model makes an autonomous judgment on the control situation.

[0012] The industrial control computer uses the LLaMA open-source large model to build a dataset fine-tuning model. The trajectory dataset is divided into a training set, a validation set, and a test set. The trajectory dataset is the obstacle avoidance data collected from the Mecanum wheel chassis, specifically including the attitude and position data of the chassis. The dataset fine-tuning model is specifically as follows: the training set is passed to the local LLaMA large model, and the learning rate, batch size, number of training epochs, training accuracy, optimizer, and distributed training settings are set. The Newton iteration method is used to adjust the model weights to minimize the error of a specific task.

[0013] The embodied thinking module is based on the local LLaMA large model and conducts dialogue and communication with the local LLaMA large model through natural language. It provides the obstacle distance information of the 2D lidar and the rotation speed information of the DC brushless motor to the local LLaMA large model in a natural language manner. The large language model combines the sensor information provided by the context and gives the operating speed of the motor, and then sends it to the embodied execution module. A human-computer interaction module is designed for the embodied thinking module of the robot to interact with people. The current operating situation of the robot and whether each hardware device is normal can be viewed through the display screen. The voice broadcast module is used to broadcast whether the task is completed and to broadcast abnormalities.

[0014] Local deployment of the embodied thinking module: An ubuntu 18.04 embedded system is built on the industrial control computer, and a deep learning environment is built on this basis. After successfully building the deep learning environment, the local LLaMA large model is locally deployed.

[0015] The display screen voice module human-computer interaction system is built based on the ROS system and is used for the embodied thinking system to interact with people. People issue commands or have conversations with the large model through the UI interface on the display screen. The natural language answered by the large model is broadcast through the voice module, and the sensor information directly subscribed back by the CAN to USB module is displayed and broadcast.

[0016] The Mecanum wheels have two models, left-handed and right-handed, and are installed in an X shape, that is, the diagonal rotation directions are the same installation configuration. The Arduino brushless motor controller is used to perform speed loop control on the DC brushless motor to achieve the in-situ rotation of the robot and movement in any direction. Then, the attitude closed-loop control of the entire robot is carried out with the IMU on the Arduino brushless motor controller. The speed information of the brushless motor is calculated and uploaded to the industrial control computer through CAN bus communication, and the speed control signal is received to control the movement of the robot.

[0017] The specific calculation of the speed information is as follows: The omnidirectional movement of the embodied execution module is achieved through the method of speed vector synthesis, and the FOC vector control is adopted in the motor control algorithm.

[0018] The beneficial effects of adopting the above technical solutions are as follows:

[0019] The present invention provides a robot autonomous navigation system based on an AI large model, and proposes a lightweight autonomous navigation framework for mobile robots. Compared with traditional autonomous navigation frameworks, it is more efficient, has a simpler process, and stronger environmental adaptability:

[0020] Traditional autonomous navigation framework: a. It is necessary to perform data fusion processing of sensors to construct a global map and obtain position information from the global map; b. Combine odometer data to obtain the pose, speed, and acceleration of the robot; c. Obtain the changing surrounding environment according to sensor data; d. After obtaining these map, pose, and speed data, use a global planning algorithm to plan a global path, e. Then use a local path planning algorithm to perform secondary planning in the changing surrounding environment, f. After planning the local path, use path optimization to obtain the executable trajectory of the robot, g. Finally, the robot actuator follows the trajectory.

[0021] The autonomous navigation framework of the present invention: a. Obtain sensor data and provide it to the large model; b. The large model generates control instructions according to the sensor data; c. The robot actuator executes the control instructions of the large model Brief Description of the Drawings

[0022] Figure 1 It is the overall structure block diagram of the robot autonomous navigation system in the embodiment of the present invention;

[0023] Figure 2 It is the overall appearance diagram of the robot autonomous navigation system in the embodiment of the present invention;

[0024] Figure 3 It is the schematic diagram of the fine-tuning process of the large module in the embodiment of the present invention;

[0025] Figure 4 It is the robot actuator diagram in the embodiment of the present invention;

[0026] Among them, (a) is the schematic diagram of the human-computer interaction module, (b) is the schematic diagram of the embodied perception module, (c) is the schematic diagram of the embodied execution module, and (d) is the schematic diagram of the interactive unified power supply system;

[0027] Figure 5 It is the structure diagram of the interactive power supply system in the embodiment of the present invention;

[0028] Figure 6 It is the block diagram of the autonomous navigation system of the AI large model in the embodiment of the present invention;

[0029] In the figure: 1 - Human - machine interaction module, 2 - Embodied perception module, 3 - Embodied execution module, 4 - Interactive unified power supply system; 101 - Display screen, 102 - Voice broadcast module, 201 - 2D lidar, 202 - Industrial control computer, 203 - CAN to USB module, 301 - Mecanum wheel, 302 - DC brushless motor, 303 - Arduino brushless motor controller; 401 - 24V 4000mah lithium battery, 402 - 24V to 19V voltage - stabilizing module, 403 - 24V to 12V voltage - stabilizing module, 404 - 24V to 5V voltage - stabilizing module, 405 - Emergency stop switch. Detailed implementation manners

[0030] The following combines the drawings and embodiments to further describe in detail the specific implementation manners of the present invention. The following embodiments are used to illustrate the present invention, but are not used to limit the scope of the present invention.

[0031] A robot autonomous navigation system based on the AI large - model, as Figure 1 shown, includes a human - machine interaction module, an embodied perception module, an embodied thinking module, an embodied execution module, and an interactive unified power supply system;

[0032] In this embodiment, as Figure 2 shown, the human - machine interaction module 1 includes a display screen 101 and a voice broadcast module 102. The embodied perception module 2 includes a 2D lidar 201 and a CAN to USB module 203. The embodied thinking module is an industrial control computer 202. The embodied execution module 3 includes a Mecanum wheel 301, a DC brushless motor 302, and an Arduino brushless motor controller 303. The interactive unified power supply system 4 includes a 24V 4000mah lithium battery 401, a 24V to 19V voltage - stabilizing module 402, a 24V to 12V voltage - stabilizing module 403, a 24V to 5V voltage - stabilizing module 404, and an emergency stop switch 405; The robot actuator diagram is as Figure 4 shown, where Figure 4 (a) is a schematic diagram of the human - machine interaction module, Figure 4 (b) is a schematic diagram of the embodied perception module, Figure 4 (c) is a schematic diagram of the embodied execution module, Figure 4 (d) is a schematic diagram of the interactive unified power supply system;

[0033] The embodied perception module 2 obtains the surrounding environment information through the 2D lidar 201, mainly the distance information from the robot to nearby obstacles. The industrial control computer 202 performs coordinate transformation on the data obtained by the 2D lidar (201), converting the obstacle distance information in the radar coordinate system into the obstacle distance information in the robot coordinate system. The industrial control computer 202 has a locally deployed LLaMA large model under the ubuntu system, and provides the obstacle distance information to the LLaMA local large model to achieve end-to-end autonomous navigation planning for the robot;

[0034] In the embodied execution module 3, the Mecanum wheels 301 are used to achieve omnidirectional movement of the robot, enabling the robot to have higher movement flexibility and be able to move in any posture and orientation on the ground; the brushless DC motor 302 is used to provide power for the chassis; the Arduino brushless motor controller 303 performs FOC vector control on the brushless DC motor 302 so that the motor can obtain a more accurate rotation speed. Arduino communicates with the industrial control computer through the CAN to USB module 203 and receives the speed control signal from the industrial control computer to control the movement of the chassis.

[0035] The interactive unified power supply system 4 is used to supply power to the entire system. The 24V 4000mah lithium battery 401 is used to supply power to the entire system, directly supply power to the brushless DC motor 302. The 24V 4000mah lithium battery 401 supplies power to the industrial control computer 202 through the 24V to 19V voltage stabilization module 402, and the industrial control computer 202 is responsible for supplying power to the display screen 101 and the voice broadcast module 102; it supplies power to the 2D lidar 201 through the 24V to 12V voltage stabilization module 403; it supplies power to the Arduino brushless motor controller 303 through the 24V to 5V voltage stabilization module 404. The emergency stop switch 405 is used for starting and stopping the power supply of the entire system.

[0036] The embodied perception module is mainly based on a lidar sensor, as Figure 2 shown.

[0037] The 2D lidar 201 used in this embodiment is the Beiyang UST-30LX lidar. This lidar has a measurement range of 0.1 to 30m, a measurement distance accuracy of ±40mm; a measurement angle range of 270° and an angle resolution of up to 0.25°; a lidar point cloud update frequency of 40Hz; the processor model of the industrial control computer is the 8th generation I5 without GPU, with a memory of 16G and a storage of 128G, and it uses the ubuntu18.04 embedded system.

[0038] The 2D lidar 201 communicates with the industrial computer 202 via the USB2.0 protocol. The industrial computer 202 performs coordinate transformation according to the set relative position relationship between the 2D lidar 201 and the chassis of the Mecanum wheel 301, converts the sensor data of the 2D lidar 201 into the sensor data of the robot itself, and provides it to the local LLaMA large model for the interaction between the large model and the external environment.

[0039] The local LLaMA large model controls the posture of the robot according to the obstacle distance information transmitted back by the 2D lidar 201. The industrial computer 202 sends the speed command to the Arduino brushless motor controller 303 through the CAN to USB module 203. The Arduino brushless motor controller 303 sends the posture data of the chassis and the feedback data of the DC brushless motor 302 to the industrial computer 202, and the local LLaMA large model makes an autonomous judgment on the control situation.

[0040] The industrial computer 202 uses the LLaMA open-source large model to construct a dataset fine-tuning model. LLaMA (Large Language Model Meta AI) is a large model developed by Meta AI and is a large-scale model based on the Transformer architecture. According to the robot autonomous navigation system see Figure 1 , the trajectory dataset is divided into a training set, a validation set, and a test set. The trajectory dataset is the obstacle avoidance data of the Mecanum wheel 301 chassis collected, specifically including the posture and position data of the chassis. The dataset fine-tuning model is specifically to transfer the training set to the local LLaMA large model, set the learning rate, batch size, number of training epochs, training accuracy, optimizer, distributed training settings, and use the Newton iteration method to adjust the model weights to minimize the error of a specific task, such as Figure 3 shown.

[0041] The embodied thinking module is based on the local LLaMA large model, conducts dialogue and communication with the local LLaMA large model through natural language, provides the obstacle distance information of the 2D lidar 201 and the rotation speed information of the DC brushless motor 302 to the local LLaMA large model in a natural language manner. The large language model combines the sensor information provided by the context and gives the operating speed of the motor, and then sends it to the embodied execution module to achieve precise motion control of the chassis. The human-computer interaction module 1 is designed for the interaction between the embodied thinking module of the robot and the personnel. The current operating situation of the robot and whether each hardware device is normal can be viewed through the display screen 101, and the voice broadcast module 102 is used to broadcast whether the task is completed and abnormal broadcasts.

[0042] Local deployment of the embodied thinking module: An ubuntu 18.04 embedded system was set up on the industrial control computer 202, and a deep learning environment was built on this basis. Since the embodied thinking module is based on the open-source large model LLaMA, and the operation of the large model depends on the deep learning environment, the LLaMA local large model was locally deployed after successfully building the deep learning environment. And a C++ programming and development environment under the embedded system was deployed because it is necessary to communicate with the large model through C++ programming. The advantage of locally deploying the large language model is that it is more secure, data will not leak, there will be no interaction with the network, and the robot will not be interfered by the network to produce dangerous behaviors that affect the safety of personnel.

[0043] Build a local embodied thinking system. Based on the ROS robot operating system, use various algorithm APIs provided by ROS for robot control, such as coordinate transformation algorithms and various driver packages for communication with the lower computer. Use C++ for local programming and communicate with the lower computer through the CAN to USB module to obtain various sensor information of the robot, and publish it in the form of a topic through the TCP protocol. The large model subscribes to various sensor information in the ROS system, forms natural language and then interacts with the large model. The large model makes control instructions for the robot according to the sensor information and publishes them again using the communication protocol provided by ROS. The above is the process of building the local embodied thinking system of the present invention, and then it is transmitted to the lower computer through the CAN to USB module for control.

[0044] The human-computer interaction system of the display screen voice module is built based on the ROS system and is used for the embodied thinking system to interact with personnel. Personnel issue commands or have conversations with the large model through the UI interface on the display screen, and the natural language answered by the large model is broadcast through the voice module. To ensure real-time performance, the monitoring of the hardware system does not pass through the large language model, but directly subscribes to the sensor information returned by the CAN to USB module 203 for display and broadcast. The block diagram of the autonomous navigation system based on the AI large model in this embodiment is as Figure 6 shown.

[0045] The embodied execution module is the Mecanum wheel 301 mobile system, as Figure 4 shown.

[0046] The Mecanum wheel 301 has two models, left-handed and right-handed, and is installed in an X shape, that is, the diagonal rotation directions are the same installation configuration, as Figure 4The installation method shown in the figure. The Arduino brushless motor controller 303 is used to perform accurate and rapid speed loop control on the DC brushless motor 302, enabling the robot to rotate in place and move in any direction. In this embodiment, there are four DC brushless motors 302, which respectively control four Mecanum wheels, and then perform attitude closed-loop control of the entire robot with the IMU on the Arduino brushless motor controller 303. After calculating the speed information of the brushless motor, it is uploaded to the industrial computer through CAN bus communication, and the speed control signal is received to control the movement of the robot;

[0047] The premise for the Mecanum wheels to achieve omnidirectional movement is to accurately control the speed of the motor. The specific calculation of the speed information is as follows: The omnidirectional movement of the embodied execution module is achieved through the method of speed vector synthesis, and the FOC vector control is adopted in the motor control algorithm.

[0048] The Arduino control board used by the industrial computer 202 has a hardware CAN bus communication interface, and CAN is converted to USB to communicate with the host computer at high speed and stability. The communication protocol between the upper and lower computers is formulated independently to achieve high-efficiency transmission of data. When communicating, the upper and lower computers verify the data through the CRC algorithm to ensure the correctness of the data and prevent data distortion caused by external interference.

[0049] In this embodiment, the interactive unified power supply system 4 is as Figure 5 shown. Many of the devices involved in the present invention require power supply, including brushless motors, Arduino control boards, industrial computers, lidar, displays, voice broadcast modules, and CAN to USB modules. The power supply of the brushless motor is 24V, and a 24V lithium battery can be directly used for power supply; the power supplies of the Arduino control board, display, voice broadcast module, and CAN to USB module are all 5V, and the 24V battery needs to be step-down and voltage-stabilized; the power supply voltage of the industrial computer is 19V, and step-down and voltage-stabilization are also required.

[0050] The Arduino control board brushless motor Hall sensor and current sensor are powered by a 24V to 5V 6A voltage stabilization module. The CAN to USB module is powered by the USB port of the industrial computer, and the lithium battery directly powers the brushless motor. The voltage stabilization modules used all adopt electrical isolation to prevent the back electromotive force and current interference of the motor coil from affecting the sensor system and control system.

[0051] The industrial control computer is powered by a voltage stabilizing module that converts 24V to 19V 6A. The rated power of the industrial control computer is 95W. Since the industrial control computer needs to supply power to the voice broadcast module, the display screen, and the CAN to USB module, the power of the voltage stabilizing module needs to have redundancy to ensure the normal operation of the system. The voltage stabilizing module adopted here also has electrical isolation; the lidar is powered by a voltage stabilizing module that converts 24V to 12V 2A. The lidar is a precision sensing device, so this module has been specially processed and has better electrical isolation effect.

[0052] The above description is only a preferred embodiment of the present disclosure and an explanation of the applied technical principles. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by the specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above inventive concept. For example, the technical solutions formed by mutually replacing the above features with the technical features (but not limited to) disclosed in the embodiments of the present disclosure that have similar functions.

Claims

1. A robot autonomous navigation system based on an AI large model, characterized in that, It includes a human-computer interaction module, an embodied perception module, an embodied thinking module, an embodied execution module, and an interactive unified power supply system; The human-computer interaction module (1) includes a display screen (101) and a voice broadcast module (102). The embodied perception module (2) includes a 2D lidar (201) and a CAN to USB module (203). The embodied thinking module is an industrial control computer (202). The embodied execution module (3) includes Mecanum wheels (301), a brushless DC motor (302), and an Arduino brushless motor controller (303). The interactive unified power supply system (4) includes a 24V 4000mAh lithium battery (401), a 24V to 19V voltage stabilization module (402), a 24V to 12V voltage stabilization module (403), a 24V to 5V voltage stabilization module (404), and an emergency stop switch (405); The embodied perception module (2) obtains surrounding environment information through the 2D lidar (201). The industrial control computer (202) performs coordinate transformation on the data obtained by the 2D lidar (201), converting the obstacle distance information in the radar coordinate system into the obstacle distance information in the robot coordinate system. The industrial control computer (202) has a local LLaMA large model arranged under the ubuntu system, and provides the obstacle distance information to the local LLaMA large model to achieve end-to-end autonomous navigation planning of the robot; In the embodied execution module (3), the Mecanum wheels (301) are used to achieve the omnidirectional movement of the robot, and the brushless DC motor (302) is used to provide power for the chassis. The Arduino brushless motor controller (303) performs FOC vector control on the brushless DC motor (302). Arduino communicates with the industrial control computer through the CAN to USB module (203) and receives the speed control signal from the industrial control computer to control the movement of the chassis; The interactive unified power supply system (4) is used to supply power to the entire system. The 24V 4000mAh lithium battery (401) is used to supply power to the entire system and directly supply power to the brushless DC motor (302). The 24V 4000mAh lithium battery (401) supplies power to the industrial control computer (202) through the 24V to 19V voltage stabilization module (402), and the industrial control computer (202) is responsible for supplying power to the display screen (101) and the voice broadcast module (102); it supplies power to the 2D lidar (201) through the 24V to 12V voltage stabilization module (403); it supplies power to the Arduino brushless motor controller (303) through the 24V to 5V voltage stabilization module (404). The emergency stop switch (405) is used to start and stop the power supply of the entire system.

2. The robot autonomous navigation system based on the AI large model according to claim 1, wherein The 2D lidar (201) communicates with the industrial control computer (202) through the USB2.0 protocol. The industrial control computer (202) performs coordinate transformation according to the relative position relationship between the set 2D lidar (201) and the chassis of the Mecanum wheel (301), converts the sensor data of the 2D lidar (201) into the sensor data of the robot itself, and provides it to the local LLaMA large model for the interaction between the large model and the external environment.

3. The robot autonomous navigation system based on the AI large model according to claim 1, characterized in that, The local deployment of the embodied thinking module is specifically as follows: An ubuntu 18.04 embedded system is built on the industrial control computer (202), and a deep learning environment is built on this basis. After successfully building the deep learning environment, the local LLaMA large model is deployed locally.

4. The robot autonomous navigation system based on the AI large model according to claim 3, characterized in that, The local LLaMA large model controls the posture of the robot according to the obstacle distance information transmitted back by the 2D lidar (201). The industrial control computer (202) sends the speed command to the Arduino brushless motor controller (303) through the CAN to USB module (203). The Arduino brushless motor controller (303) sends the posture data of the chassis and the feedback data of the DC brushless motor (302) to the industrial control computer (202), and the local LLaMA large model makes an independent judgment on the control situation; The industrial control computer (202) constructs a dataset fine-tuning model using the LLaMA open-source large model, and divides the trajectory dataset into a training set, a validation set, and a test set. The trajectory dataset is the obstacle avoidance data of the Mecanum wheel (301) chassis collected, specifically including the posture and position data of the chassis; The dataset fine-tuning model is specifically to pass the training set to the local LLaMA large model, set the learning rate, batch size, number of training epochs, training accuracy, optimizer, distributed training settings, and use the Newton iteration method to adjust the model weights to minimize the error of a specific task; The embodied thinking module is based on the local LLaMA large model and conducts dialogue and communication with the local LLaMA large model through natural language. It provides the obstacle distance information of the 2D lidar (201) and the rotation speed information of the DC brushless motor (302) to the local LLaMA large model in a natural language manner. The large language model provides the running speed of the motor in combination with the sensor information provided in the context, and then sends it to the embodied execution module. The human-computer interaction module (1) is designed for the interaction between the embodied thinking module of the robot and the personnel. The current running situation of the robot and whether each hardware device is normal can be viewed through the display screen (101), and the voice broadcast module (102) is used to broadcast whether the task is completed and abnormal broadcasts.

5. A robot autonomous navigation system based on an AI large model according to claim 1, characterized in that, The human-computer interaction module is built based on the ROS system for the interaction between the embodied thinking system and the personnel. The personnel issue commands or have conversations with the large model through the UI interface on the display screen, and the natural language answered by the large model is broadcast through the voice module. It directly subscribes to the sensor information transmitted back by the CAN to USB module (203) for display and broadcast.

6. The robot autonomous navigation system based on the AI large model according to claim 1, characterized in that, The Mecanum wheel (301) has two models, left-handed and right-handed, and is installed in an X shape, that is, the diagonal rotation directions are the same installation configuration. The Arduino brushless motor controller (303) is used to perform speed loop control on the DC brushless motor (302) to enable the robot to rotate in place and move in any direction. Then, the IMU on the Arduino brushless motor controller (303) is used for the attitude closed-loop control of the entire robot. After the speed information of the brushless motor is calculated, it is uploaded to the industrial control computer through CAN bus communication, and the speed control signal is received to control the movement of the robot.

7. An autonomous navigation system for a robot based on an AI large model according to claim 6, characterized in that, The specific calculation of the speed information is as follows: The omnidirectional movement of the embodied execution module is realized through the method of speed vector synthesis, and the FOC vector control is adopted in the motor control algorithm.

Images