Intelligent robot end-to-end motion planning method

Abstract

The invention discloses an intelligent robot end-to-end motion planning method. The method comprises the following steps: constructing a simulation environment for extraterrestrial detection; the simulation environment is used for simulating motion characteristics of the robot in the extraterrestrial detection environment and wheel soil contact dynamic characteristics, and outputting sensing information of the robot based on a received action instruction; an end-to-end motion planning model is constructed, and the motion planning model takes perception information of the robot as input and takes the action instruction as output; the motion planning model is provided with a reward function, and the reward function is fused with artificial priori knowledge and actual engineering constraints; training the motion planning model based on a simulation environment to obtain a trained model; and in response to deployment of the trained model to an upper computer of the extraterrestrial exploration robot, enabling the robot to output a motion planning result based on real sensing information measured by a sensor of the robot. According to the invention, computing resources are saved, and the end-to-end motion planning capability of the robot under the constraint of limited resources can be improved.

Description

Technical Field

[0001] The present invention relates to the field of extraterrestrial exploration technology, and in particular to an end-to-end motion planning method for an intelligent robot. Background Art

[0002] Due to the limitations of deep space exploration, such as harsh and unknown environment, lack of prior knowledge, limited computing resources, and long signal transmission distance, the extraterrestrial exploration system needs to have autonomous motion planning capabilities to improve the efficiency of extraterrestrial exploration and meet the real-time and robustness requirements of exploration.

[0003] At present, the existing end-to-end motion planning methods rely on the deep reinforcement learning framework, which can achieve direct mapping from sensor data input to motion control command output. However, the existing methods need to rely on large-scale computing resources and a large number of labeled data sets for training. The computing resources are high and the data sample volume is large, which is difficult to meet the conditions of limited resources and lack of samples in extraterrestrial exploration.

[0004] Therefore, there is an urgent need for an intelligent robot end-to-end motion planning method to solve the above problems. Summary of the invention

[0005] The present invention provides an end-to-end motion planning method for an intelligent robot, which has low requirements on computing resources and data sample volume, and can improve the end-to-end motion planning capability of an extraterrestrial exploration mobile robot under limited resource constraints. The technical solution is as follows:

[0006] On the one hand, an end-to-end motion planning method for an intelligent robot is provided, the method comprising:

[0007] Constructing a simulation environment for extraterrestrial exploration; the simulation environment is used to simulate the motion characteristics and wheel-soil contact dynamics of the robot in the extraterrestrial exploration environment, and output the robot's perception information based on the received action instructions;

[0008] Construct an end-to-end motion planning model, wherein the motion planning model takes the robot's perception information as input and takes action instructions as output; and the motion planning model is designed with a reward function, which integrates artificial prior knowledge and actual engineering constraints;

[0009] Training the motion planning model based on the simulation environment to obtain a trained model;

[0010] In response, the trained model is deployed to a host computer of an extraterrestrial exploration robot so that the robot outputs motion planning results based on real perception information measured by its sensors.

[0011] On the other hand, an end-to-end motion planning device for an intelligent robot is provided, the device comprising:

[0012] A first construction unit is used to construct a simulation environment for extraterrestrial exploration; the simulation environment is used to simulate the motion characteristics and wheel-soil contact dynamics characteristics of the robot in the extraterrestrial exploration environment, and output the perception information of the robot based on the received action instructions;

[0013] The second construction unit is used to construct an end-to-end motion planning model, wherein the motion planning model takes the perception information of the robot as input and takes the action command as output; and the motion planning model is designed with a reward function, and the reward function integrates artificial prior knowledge and actual engineering constraints;

[0014] A training unit, used for training the motion planning model based on the simulation environment to obtain a trained model;

[0015] A planning unit is used to respond to deploying the trained model to a host computer of an extraterrestrial exploration robot so that the robot outputs motion planning results based on real perception information measured by its sensors.

[0016] On the other hand, a computer-readable storage medium is provided, in which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned intelligent robot end-to-end motion planning method are implemented.

[0017] On the other hand, a computer program product is provided, comprising a computer program, which implements the steps of the above-mentioned intelligent robot end-to-end motion planning method when executed by a processor.

[0018] An embodiment of the present invention provides an end-to-end motion planning method for an intelligent robot. First, the constructed simulation environment can model the movement of a robot in an extraterrestrial detection environment and output simulation data. The motion planning model is trained using the simulation data, which can reduce the need for real data samples such as manual annotation. Secondly, by integrating artificial prior knowledge and actual engineering constraints into the design of the reward function, and performing model training based on the reward function, the learning efficiency of the model can be improved, the model can be accelerated to learn in a direction that meets the requirements, unnecessary exploration can be avoided, and the demand for computing resources and data samples can be reduced. Finally, the trained model is deployed in the robot, which can realize end-to-end autonomous planning of the robot and autonomously adjust the driving speed and steering speed of the robot's wheels. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0020] Figure 1 This is a flow chart of an end-to-end motion planning method for an intelligent robot provided by one embodiment of the present invention;

[0021] Figure 2 It is a structural diagram of an end-to-end motion planning device for an intelligent robot provided by one embodiment of the present invention;

[0022] Figure 3 is a hardware architecture diagram of a computer device provided by an embodiment of the present invention;

[0023] Figure 4 Schematic diagram of a network of a motion planning model provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0025] The specific implementation of the above concept is described below.

[0026] Please refer to Figure 1 , an embodiment of the present invention provides an end-to-end motion planning method for an intelligent robot, the method comprising:

[0027] Step 100, constructing a simulation environment for extraterrestrial exploration; the simulation environment is used to simulate the motion characteristics and wheel-soil contact dynamics characteristics of the robot in the extraterrestrial exploration environment, and output the robot's perception information based on the received action instructions;

[0028] Step 102, constructing an end-to-end motion planning model, the motion planning model takes the robot's perception information as input and takes action instructions as output; and the motion planning model is designed with a reward function, which integrates artificial prior knowledge and actual engineering constraints;

[0029] Step 104, training the motion planning model based on the simulation environment to obtain a trained model;

[0030] Step 106, in response to deploying the trained model to the host computer of the extraterrestrial exploration robot, so that the robot outputs motion planning results based on the real perception information measured by its sensors.

[0031] In this embodiment, first, the constructed simulation environment can model the movement of the robot in the extraterrestrial detection environment and output simulation data. The motion planning model is trained using the simulation data, which can reduce the need for real data samples such as manual annotation. Secondly, by integrating artificial prior knowledge and actual engineering constraints into the design of the reward function, and training the model based on the reward function, the learning efficiency of the model can be improved, the model can be accelerated to learn in a direction that meets the requirements, unnecessary exploration can be avoided, and the demand for computing resources and data samples can be reduced. Finally, the trained model is deployed in the robot, which can realize the end-to-end autonomous planning of the robot and autonomously adjust the driving speed and steering speed of the robot's wheels.

[0032] Described below Figure 1 How the various steps are performed.

[0033] With respect to step 100, a simulation environment for extraterrestrial exploration is constructed.

[0034] In this step, the simulation environment is built based on AGX Dynamics technology. The simulation model includes a robot model and a soil model agxTerrain; the soil model is used to simulate the dynamic interaction and deformation between the robot's wheels and the soil, and is designed with soil-related parameters and the physical state of the soil, including soft soil settlement depth, soil compaction, and wheel-soil friction coefficient. The soil model is shown in Table 1:

[0035] Table 1 Soil model parameters

[0036]

[0037] The robot model is designed with wheel speed and steering control interfaces to receive action instructions output by the motion planning model based on the interface and perform corresponding actions based on the action instructions. In addition, the robot model also reserves interfaces for end-to-end control of subsequent motion planning model learning, training and reasoning verification.

[0038] For step 102, an end-to-end motion planning model is constructed.

[0039] The motion planning model is designed with a reward function, which includes intermediate rewards, sparse rewards and constrained rewards;

[0040] The expression of the intermediate reward is:

[0041] Intermediate Reward=-b+a×(v t 2 -v t-1 2 )

[0042] The expression of sparse reward is:

[0043] Sparse Reward = c

[0044] The expression of constraint reward is:

[0045] Constraint Reward=-d×|A i -A iconstraint |

[0046] In the formula, Intermediate Reward represents the intermediate reward; Sparse Reward represents the sparse reward; Constraint Reward represents the constraint reward; a is the intermediate reward coefficient, which is used to reward the increase in the robot's body speed; b is the negative feedback constant term, which is used to speed up the convergence of the motion planning model; c is the sparse reward constant, which is used to reward the robot for successfully escaping; d is the constraint reward coefficient; A i is the value of the robot's i-th state parameter, A iconstraint is the range of the robot’s i-th state parameter; v t is the speed of the robot at time t; v t-1 is the speed of the robot at time t-1.

[0047] In this step, by designing the intermediate reward coefficient a, the increase in the vehicle speed can be rewarded. A certain positive reward can be given for each increase in speed to encourage the agent (such as a robot with the motion planning model deployed inside) to take actions that can increase the speed. By designing the negative feedback constant term b, the convergence speed can be accelerated, prompting the agent to avoid invalid behaviors and learn effective escape strategies more quickly. By setting a small negative value, there will be a slight penalty for each step, which helps to balance the exploration and utilization of the agent during the learning process and ensure that the agent does not take too many risks or deviate from the goal. By designing a sparse reward constant c, the agent can be rewarded for successfully escaping from a predicament. By setting a large positive value, the motivation of the agent to learn the final goal can be significantly enhanced. By designing the constraint reward coefficient d, when the agent control exceeds the constraint range, it will be punished to encourage the agent to learn in the direction that meets the engineering constraints. By setting a relatively small negative value, the agent will be punished accordingly when it violates the constraint conditions, balancing exploration and utilization while guiding the agent to learn in a safe and feasible direction. It should be noted that the specific values ​​of the four parameters a, b, c, and d can be determined based on engineering experience and are not specifically limited in this application.

[0048] By designing the above parameters, this step can integrate artificial prior knowledge and actual engineering constraints (i.e., rule constraints) into the design of the reward function. By conducting data-driven learning and training in a simulation environment that can simulate the real environment, not only can the effective fusion of knowledge and data be achieved, but also the expected benefits of the motion planning model output can be maximized under resource-constrained conditions, while strictly complying with engineering feasibility constraints and in line with actual engineering experience.

[0049] In some embodiments, the robot's perception information includes: wheel drive motor speed, wheel steering motor speed, wheel steering angle and vehicle body speed, wheel sinking amount, wheel slip rate and wheel-soil contact force.

[0050] In this step, the robot's perception information can be measured or calculated by its sensors.

[0051] In some implementations, the motion instructions include: rotation speed instructions and steering speed instructions for each wheel of the robot.

[0052] In this step, the action instructions use discrete action instructions, such as (-1, 0, 1). Among them, 1 represents the forward speed, 0 represents stop, and -1 represents the reverse speed. Each wheel will receive two action instructions, including the rotation speed instruction of the wheel drive motor and the steering speed instruction of the wheel steering motor. When the mobile robot executes a specific instruction, the above discrete action instructions are multiplied by the set corresponding wheel drive motor speed and steering motor speed.

[0053] It should also be noted that, in some embodiments, Figure 4 As shown, the motion planning model includes an encoder, an aggregator and a decoder arranged in sequence;

[0054] The encoder consists of two fully connected layers, a Transformer Block layer, and a maximum pooling layer. The encoder is used to receive the robot's perception information and perform feature encoding on the received information.

[0055] The aggregator consists of a 256-node fully connected layer and a 128-dimensional autoregressive component; the aggregator is used to aggregate the features output by the encoder;

[0056] The decoder includes a decision network and an evaluation network. The decision network contains a fully connected layer and a Softmax transformation layer, and the evaluation network contains two fully connected layers. The evaluation network decodes the features output by the decoder and outputs a value assessment of the current state to guide the optimization of the decision network based on the evaluation results. The decision network is used to output action instructions.

[0057] In this step, the motion planning model adopts the Transformer-based Encoder-Decoder architecture, and uses a fully connected network, Transformer Block, a maximum pooling layer, etc. to form a neural network model. The input of the neural network comes from the robot's sensory information. After the model extracts, expresses, and infers and calculates the sensory information, the output is the robot's action instructions, i.e., the motion planning instructions. It can be seen that the motion planning model of the present application does not need to separate and independently implement the links of perception, decision-making, planning, and control, but instead realizes a direct mapping from the sensor data input to the motion control command output to complete end-to-end planning.

[0058] With respect to step 104, the motion planning model is trained based on the simulation environment to obtain a trained model, including:

[0059] Get the current perception information of the robot output by the simulation environment in the current training round;

[0060] Inputting the current perception information of the robot into the motion planning model to process the current perception information of the robot based on the reward function and generate corresponding action instructions and rewards;

[0061] Based on the reward, the parameters of the motion planning model are updated, and the generated action instructions are applied to the simulation environment to obtain new perception information of the robot;

[0062] The robot's new perception information is used to train the updated motion planning model for the next round, and the cycle is repeated until a trained model is obtained.

[0063] In this step, the motion planning model performs deep reinforcement learning training on the generated perception information and rewards based on the PPO reinforcement learning algorithm, continuously updates the network parameters, and enables the robot to learn more effective strategies, generate and continuously update the model.

[0064] Finally, for step 108, after the motion planning model is trained, it can be deployed to a real extraterrestrial exploration robot. In actual exploration, the robot sensor can measure data in real time, and the robot outputs action instructions based on the measured data to complete motion planning.

[0065] like Figure 2 , Figure 3 As shown, an embodiment of the present invention provides an end-to-end motion planning device for an intelligent robot. The device embodiment can be implemented by software, or by hardware or a combination of software and hardware. From the hardware level, Figure 2 As shown in FIG. 1 , a hardware architecture diagram of a computing device where an end-to-end motion planning device for an intelligent robot provided in an embodiment of the present invention is located is shown in FIG. Figure 2In addition to the processor, memory, network interface, and non-volatile memory shown in the figure, the computing device in which the device is located in the embodiment may also generally include other hardware, such as a forwarding chip responsible for processing messages, etc. Taking software implementation as an example, Figure 3 As shown, as a device in a logical sense, the CPU of the computing device in which it is located reads the corresponding computer program in the non-volatile memory into the internal memory and runs it.

[0066] Please refer to Figure 3 , an embodiment of the present invention provides an end-to-end motion planning device for an intelligent robot, the device comprising:

[0067] The first construction unit 300 is used to construct a simulation environment for extraterrestrial exploration; the simulation environment is used to simulate the motion characteristics and wheel-soil contact dynamics characteristics of the robot in the extraterrestrial exploration environment, and output the perception information of the robot based on the received action instructions;

[0068] The second construction unit 302 is used to construct an end-to-end motion planning model, which takes the robot's perception information as input and action instructions as output; and the motion planning model is designed with a reward function, which integrates artificial prior knowledge and actual engineering constraints;

[0069] A training unit 304 is used to train the motion planning model based on the simulation environment to obtain a trained model;

[0070] The planning unit 306 is used to respond to the deployment of the trained model to the host computer of the extraterrestrial exploration robot so that the robot outputs motion planning results based on the real perception information measured by its sensors.

[0071] In some embodiments, the simulation model includes a robot model and a soil model; the soil model is used to simulate the dynamic interaction and deformation between the robot's wheels and the soil, and is designed with relevant soil parameters and the physical state of the soil, including soft soil settlement depth, soil compaction and wheel-soil friction coefficient; the robot model is designed with a wheel speed and steering control interface to receive action instructions output by the motion planning model based on the interface, and perform corresponding actions based on the action instructions.

[0072] In some implementations, the reward function includes an intermediate reward, a sparse reward, and a constrained reward;

[0073] The expression of the intermediate reward is:

[0074] Intermediate Reward=-b+a×(v t 2 -v t-1 2 )

[0075] The expression of sparse reward is:

[0076] Sparse Reward = c

[0077] The expression of constraint reward is:

[0078] Constraint Reward=-d×|A i -A iconstraint |

[0079] In the formula, Intermediate Reward represents the intermediate reward; Sparse Reward represents the sparse reward; Constraint Reward represents the constraint reward; a is the intermediate reward coefficient, which is used to reward the increase in the robot's body speed; b is the negative feedback constant term, which is used to speed up the convergence of the motion planning model; c is the sparse reward constant, which is used to reward the robot for successfully escaping; d is the constraint reward coefficient; A i is the value of the robot's i-th state parameter, A iconstraint is the range of the robot's i-th state parameter; v t is the speed of the robot at time t; v t-1 is the speed of the robot at time t-1.

[0080] In some embodiments, the robot's perception information includes: wheel drive motor speed, wheel steering motor speed, wheel steering angle and vehicle body speed, wheel sinking amount, wheel slip rate and wheel-soil contact force.

[0081] In some implementations, the motion instructions include: rotation speed instructions and steering speed instructions for each wheel of the robot.

[0082] In some implementations, the training unit 304 is configured to perform the following operations:

[0083] Inputting the current perception information of the robot into the motion planning model to process the current perception information of the robot based on the reward function and generate corresponding action instructions and rewards;

[0084] Based on the reward, the parameters of the motion planning model are updated, and the generated action instructions are applied to the simulation environment to obtain new perception information of the robot;

[0085] The robot's new perception information is used to train the updated motion planning model for the next round, and the cycle is repeated until a trained model is obtained.

[0086] In some embodiments, the motion planning model includes an encoder, an aggregator, and a decoder arranged in sequence;

[0087] The encoder consists of two fully connected layers, a Transformer Block layer, and a maximum pooling layer. The encoder is used to receive the robot's perception information and perform feature encoding on the received information.

[0088] The aggregator consists of a 256-node fully connected layer and a 128-dimensional autoregressive component; the aggregator is used to aggregate the features output by the encoder;

[0089] The decoder includes a decision network and an evaluation network. The decision network contains a fully connected layer and a Softmax transformation layer, and the evaluation network contains two fully connected layers. The evaluation network decodes the features output by the decoder and outputs a value assessment of the current state to guide the optimization of the decision network based on the evaluation results. The decision network is used to output action instructions.

[0090] It should be noted that the end-to-end motion planning device for an intelligent robot provided in the above embodiment is only illustrated by the division of the above functional modules. In practical applications, the above functional distribution can be completed by different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the end-to-end motion planning device for an intelligent robot provided in the above embodiment and the end-to-end motion planning method embodiment of an intelligent robot belong to the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0091] The embodiment of the present application also provides a computer device, please refer to Figure 3 The computer device includes a processor and a memory, in which at least one instruction, at least one program, a code set or an instruction set is stored, and the at least one instruction, at least one program, a code set or an instruction set is loaded and executed by the processor to implement the end-to-end motion planning method of the intelligent robot provided by the above-mentioned method embodiments.

[0092] An embodiment of the present application also provides a computer-readable storage medium, on which is stored at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by a processor to implement the end-to-end motion planning method for an intelligent robot provided by the above-mentioned method embodiments.

[0093] An embodiment of the present application also provides a computer program product, which includes a computer program. A processor of a computer device reads the computer program from a computer-readable storage medium, and the processor executes the computer program, so that the computer device executes any of the intelligent robot end-to-end motion planning methods in the above embodiments.

[0094] For the convenience of description, the above system or device is described by dividing it into various modules or units according to its functions. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0095] It can be known from the description of the above implementation methods that those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present application can be essentially or partly contributed to the prior art in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments of the present application or certain parts of the embodiments.

[0096] Finally, it should be noted that, in this article, relational terms such as first, second, third and fourth are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0097] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. An end-to-end motion planning method for an intelligent robot, characterized in that: The method comprises: Constructing a simulation environment for extraterrestrial exploration; the simulation environment is used to simulate the motion characteristics and wheel-soil contact dynamics of the robot in the extraterrestrial exploration environment, and output the robot's perception information based on the received action instructions; Construct an end-to-end motion planning model, wherein the motion planning model takes the robot's perception information as input and takes action instructions as output; and the motion planning model is designed with a reward function, which integrates artificial prior knowledge and actual engineering constraints; Training the motion planning model based on the simulation environment to obtain a trained model; In response, the trained model is deployed to a host computer of an extraterrestrial exploration robot so that the robot outputs motion planning results based on real perception information measured by its sensors.

2. The method according to claim 1, characterized in that The simulation model includes a robot model and a soil model; the soil model is used to simulate the dynamic interaction and deformation between the wheels of the robot and the soil, and is designed with relevant parameters of the soil and the physical state of the soil, including the soft soil settlement depth, soil compaction and wheel-soil friction coefficient; The robot model is designed with a wheel speed and steering control interface to receive action instructions output by the motion planning model based on the interface and perform corresponding actions based on the action instructions.

3. The method according to claim 1, characterized in that The reward function includes an intermediate reward, a sparse reward and a constrained reward; The expression of the intermediate reward is: Intermediate Reward=-b+a×(v t 2 -v t-1 2 ) The expression of the sparse reward is: Sparse Reward = c The expression of the constraint reward is: Constraint Reward=-d×|A i -A iconstraint | In the formula, Intermediate Reward represents the intermediate reward; Sparse Reward represents the sparse reward; Constraint Reward represents the constraint reward; a is the intermediate reward coefficient, which is used to reward the increase in the robot's body speed; b is the negative feedback constant term, which is used to accelerate the convergence speed of the motion planning model; c is the sparse reward constant, which is used to reward the robot for successfully escaping; d is the constraint reward coefficient; A i is the value of the robot's i-th state parameter, A iconstraint is the range of the robot’s i-th state parameter; v t is the speed of the robot at time t; v t-1 is the speed of the robot at time t-1.

4. The method according to claim 1, characterized in that The robot's perception information includes: wheel drive motor speed, wheel steering motor speed, wheel steering angle and vehicle speed, wheel sinking, wheel slip rate and wheel-soil contact force.

5. The method according to claim 1, characterized in that The action instructions include: rotation speed instructions and steering speed instructions for each wheel of the robot.

6. The method according to claim 1, characterized in that The step of training the motion planning model based on the simulation environment to obtain a trained model includes: Obtaining the current perception information of the robot output by the simulation environment in the current training round; Inputting the current perception information of the robot into the motion planning model to process the current perception information of the robot based on the reward function and generate corresponding action instructions and rewards; Based on the reward, the parameters of the motion planning model are updated, and the generated action instructions are applied to the simulation environment to obtain new perception information of the robot; The robot's new perception information is used to train the updated motion planning model for the next round, and the cycle is repeated until a trained model is obtained.

7. The method according to claim 1, characterized in that The motion planning model includes an encoder, an aggregator and a decoder arranged in sequence; The encoder is composed of two fully connected layers, a Transformer Block layer and a maximum pooling layer; the encoder is used to receive the perception information of the robot and perform feature encoding on the received information; The aggregator is composed of a fully connected layer with 256 nodes and a 128-dimensional autoregressive component; the aggregator is used to aggregate the features output by the encoder; The decoder includes a decision network and an evaluation network. The decision network includes a fully connected layer and a Softmax transformation layer, and the evaluation network includes two fully connected layers. The evaluation network decodes the features output by the decoder and outputs a value assessment of the current state to guide the optimization of the decision network based on the assessment results. The decision network is used to output action instructions.

8. An end-to-end motion planning device for an intelligent robot, characterized in that: The device comprises: A first construction unit is used to construct a simulation environment for extraterrestrial exploration; the simulation environment is used to simulate the motion characteristics and wheel-soil contact dynamics characteristics of the robot in the extraterrestrial exploration environment, and output the perception information of the robot based on the received action instructions; The second construction unit is used to construct an end-to-end motion planning model, wherein the motion planning model takes the perception information of the robot as input and takes the action command as output; and the motion planning model is designed with a reward function, and the reward function integrates artificial prior knowledge and actual engineering constraints; A training unit, used for training the motion planning model based on the simulation environment to obtain a trained model; A planning unit is used to respond to deploying the trained model to a host computer of an extraterrestrial exploration robot so that the robot outputs motion planning results based on real perception information measured by its sensors.

9. A computer device, characterized in that: The computer device includes a memory and a processor, the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to implement the steps of any one of the methods described in claims 1-7.

10. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method described in any one of claims 1 to 7 are implemented.

Images