Robot dynamic obstacle avoidance method and device, control equipment and robot

Abstract

The invention provides a robot dynamic obstacle avoidance method and device, control equipment and a robot, and relates to the technical field of data processing.The method comprises the steps that original environment data of a target scene where the robot is located in a preset historical time period is acquired; according to the original environment data of the preset historical time period, obtaining historical track data of the dynamic target in the target scene in the preset historical time period; performing action intention prediction by adopting a preset intention prediction network to obtain action intention prediction information of the dynamic target in preset future time; adopting a preset game model of the robot for the target scene to generate action decision information of the robot in preset future time; and according to the action decision information, generating a target control instruction of the robot in the preset future time. According to the invention, the accuracy of the control instruction and the adaptability and interaction intelligence level of the robot in a complex dynamic scene are improved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and more specifically, to a method, apparatus, control device, and robot for dynamic obstacle avoidance. Background Technology

[0002] With the development of artificial intelligence and robotics, autonomous mobile robots are widely used in warehousing and logistics, medical assistance, and other scenarios. They often need to navigate autonomously in densely populated, dynamically changing open spaces. Such environments are highly uncertain, and traditional obstacle avoidance methods are difficult to meet the requirements of safety, efficiency, and natural interaction. Therefore, the coordinated movement of robots and surrounding intelligent agents in dynamic and uncertain environments has become one of the core challenges for their practical application.

[0003] Current dynamic obstacle avoidance technologies are mainly divided into three categories: reactive rule-based methods, which generate obstacle avoidance actions based on real-time sensor data, offering fast response but lacking long-term planning; optimization-based trajectory planning methods, which model constrained optimal control problems, balancing safety and smoothness; and reinforcement learning methods, which learn end-to-end obstacle avoidance strategies through neural networks.

[0004] However, existing dynamic obstacle avoidance technologies are characterized by reactive methods that lack long-term planning and are prone to unnatural behavior, optimization methods that rely on simple motion models and are prone to collisions due to prediction failures, and reinforcement learning methods that are uninterpretable and have limited generalization. Furthermore, all existing technologies treat obstacle avoidance as a one-sided avoidance and cannot achieve cooperative navigation. Summary of the Invention

[0005] The purpose of this application is to address the shortcomings of the prior art by providing a method, apparatus, control device, and robot for dynamic obstacle avoidance, thereby improving the accuracy of control commands and the robot's adaptability and interactive intelligence in complex dynamic scenarios.

[0006] To achieve the above objectives, the technical solutions adopted in the embodiments of this application are as follows: In a first aspect, one embodiment of this application provides a method for dynamic obstacle avoidance of a robot, the method comprising: Obtain raw environmental data of the target scene where the robot is located within a preset historical time period; Based on the original environmental data of the preset historical time period, obtain the historical trajectory data of the dynamic target in the target scene during the preset historical time period; Based on the historical trajectory data of the dynamic target in the preset historical time period, a preset intention prediction network is used to predict the action intention of the dynamic target in the preset future time to obtain the action intention prediction information of the dynamic target in the preset future time. Based on the predicted action intention information of the dynamic target at the preset future time, the current state of the robot, and the preset planned path of the robot for the target scenario, the robot's action decision information at the preset future time is generated using the preset game model of the robot for the target scenario. Based on the action decision information, the robot generates target control instructions for the preset future time.

[0007] Optionally, the original environmental data includes point cloud data and image data; obtaining the historical trajectory data of the dynamic target in the target scene within the preset historical time period based on the original environmental data of the preset historical time period includes: The image data and the point cloud data are fused and semantically recognized to obtain the semantic labels of each point cloud cluster in the point cloud data; Based on the semantic tags of each point cloud cluster, the historical trajectory data of the dynamic target during the preset historical time period is obtained from the point cloud data.

[0008] Optionally, the historical trajectory data includes: the position and velocity of the dynamic target at multiple historical trajectory points; The step of predicting the motion intent of the dynamic target in a preset future time by using a preset intent prediction network based on the motion data of the dynamic target in the preset historical time period includes: Based on the position and velocity of the dynamic target at multiple historical trajectory points, the intention prediction network is used to predict the action intention, thereby obtaining the action intention prediction information of the dynamic target at a preset future time. The action intention prediction information includes the probability of the dynamic target's intention to multiple preset actions.

[0009] Optionally, before generating the robot's action decision information at the preset future time based on the predicted action intention information of the dynamic target at the preset future time, the robot's current state, and the robot's preset planned path for the target scenario, and using a preset game model for the target scenario, the method further includes: A first state set of the robot and a second state set of the dynamic target are obtained, wherein the preset state set of the robot includes: multiple reachable positions of the robot in the target scene and a preset speed range of the robot in the target scene, and the second state set of the dynamic target includes: multiple reachable positions of the dynamic target in the target scene and a preset speed range of the dynamic target in the target scene; Obtain a first set of actions for the robot and a second set of actions for the dynamic target, wherein the first set of actions for the robot includes actions that the robot can perform in the target scene, and the second set of actions for the dynamic target includes actions that the dynamic target can perform in the target scene; Based on the first state set, the second state set, the first action set, and the second action set, a partially observable Markov game process model is constructed as the preset game model.

[0010] Optionally, the step of generating the robot's action decision information at the preset future time based on the predicted action intention information of the dynamic target at the preset future time, the robot's current state, and the robot's preset planned path for the target scenario, using a preset game model for the target scenario, includes: Based on the historical trajectory data of the dynamic target during the preset historical time period and the action intention prediction information for the preset future time, the robot constructs an observation set for the dynamic target. Based on the robot's current state, the preset planned path, and the observation set, the preset game model is used to solve the problem and generate the robot's action decision information at the preset future time.

[0011] Optionally, generating the target control command for the robot at the preset future time based on the action decision information includes: Based on the action decision information, a short-term trajectory for the preset future time is generated; Based on the current state, the preset planned path, and the short-term trajectory, the target control command is obtained by solving the preset dynamic constraint algorithm of the robot.

[0012] Optionally, the method further includes: If the dynamic target is a target intelligent agent, the short-term trajectory is communicated to the dynamic target.

[0013] Optionally, the step of solving the target control command using the robot's preset dynamic constraint algorithm based on the current state, the preset planned path, and the short-term trajectory includes: Based on the current state, obtain the next state of the robot at the next moment; Based on the robot's next state and the preset state corresponding to the next moment on the preset planning path, the first variable parameter is obtained; The second variable parameter is obtained based on the robot's next state and the predicted state corresponding to the next moment on the short-term trajectory; The third variable parameter is obtained based on the robot's next state and the pre-acquired next state of the dynamic target at the next moment; The fourth variable parameter is obtained based on multiple preset control commands; Based on the first variable parameter, the second variable parameter, the third variable parameter, and the fourth variable parameter, construct the objective function for solving; Based on the objective function, the preset dynamic constraint algorithm is solved to determine the optimal target control command from the plurality of preset control commands.

[0014] Secondly, another embodiment of this application provides a robot dynamic obstacle avoidance device, the device comprising: The first acquisition module is used to acquire the original environmental data of the target scene where the robot is located during a preset historical time period; The second acquisition module is used to acquire historical trajectory data of dynamic targets in the target scene during the preset historical time period based on the original environmental data of the preset historical time period. The prediction module is used to predict the action intent of the dynamic target in a preset future time by using a preset intent prediction network based on the historical trajectory data of the dynamic target in the preset historical time period. The first generation module is used to generate the robot's action decision information at the preset future time based on the predicted action intention information of the dynamic target at the preset future time, the current state of the robot, and the preset planned path of the robot for the target scenario, using a preset game model of the robot for the target scenario. The second generation module is used to generate target control instructions for the robot at the preset future time based on the action decision information.

[0015] Thirdly, another embodiment of this application provides a control device, including a processor and a memory, wherein the memory stores machine-readable instructions executable by the processor, and when the control device is running, the processor executes the machine-readable instructions to perform the steps of any of the robot dynamic obstacle avoidance methods described in the first aspect above.

[0016] Fourthly, another embodiment of this application provides a robot, which includes at least: a robot body and a control device disposed within the robot body, the control device being used to perform the steps of the robot dynamic obstacle avoidance method as described in any of the first aspects above.

[0017] Fifthly, another embodiment of this application provides a storage medium storing a computer program, which, when executed by a processor, performs the steps of the robot dynamic obstacle avoidance method described in any of the first aspects above.

[0018] The beneficial effects of this application are: A method, apparatus, control device, and robot for dynamic obstacle avoidance are disclosed. The method involves acquiring raw environmental data of the target scene where the robot is located within a preset historical time period; obtaining historical trajectory data of a dynamic target within the target scene during the preset historical time period based on the raw environmental data; predicting the action intent of the dynamic target at a preset future time using a preset intent prediction network based on the historical trajectory data of the dynamic target during the preset historical time period; generating action decision information for the robot at the preset future time using a preset game model for the target scene based on the predicted action intent information of the dynamic target at the preset future time; and generating target control commands for the robot at the preset future time based on the action decision information. This application improves the adaptability and interactive intelligence of the robot in complex dynamic scenes by integrating dynamic target intent prediction driven by historical environmental data with an action decision mechanism based on a game model. By utilizing historical trajectory data and deep learning networks to proactively predict the future behavioral intentions of other dynamic targets, and combining the robot's own state with the planned path, this application collaboratively optimizes its own action decisions within a game-theoretic framework, achieving safer, more efficient, and more predictable motion control. This application avoids local optima traps and frequent emergency braking problems, and improves the robustness of path generation. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart illustrating a robot dynamic obstacle avoidance method provided in an embodiment of this application; Figure 2 A flowchart illustrating the determination of historical trajectory data in a robot dynamic obstacle avoidance method provided in this application; Figure 3 A flowchart illustrating the process of determining a preset game model for a robot dynamic obstacle avoidance method provided in this application embodiment; Figure 4A flowchart illustrating the determination of action decision information in a robot obstacle avoidance method provided in this application embodiment; Figure 5 A flowchart illustrating the generation of target control commands in a robot dynamic obstacle avoidance method provided in this application embodiment; Figure 6 A schematic diagram illustrating another process for determining target control commands in dynamic obstacle avoidance of a robot, provided as an embodiment of this application; Figure 7 This is a schematic diagram of the structure of a robot dynamic obstacle avoidance device provided in an embodiment of this application; Figure 8 A schematic diagram of a control device structure provided in an embodiment of this application; Figure 9 A robot provided in an embodiment of this application. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0022] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0023] It should be noted that the term "comprising" will be used in the embodiments of this application to indicate the presence of the features declared thereafter, but does not exclude the addition of other features.

[0024] Currently, in the process of robot obstacle avoidance, only a corresponding scene model can be constructed based on the target scene. The robot's path is then planned based on the position of each dynamic target in the scene model to obtain the obstacle avoidance path. However, the process of predicting the trajectory of dynamic targets in the target scene is not involved. As a result, the robot may collide with dynamic targets during its movement based on the generated obstacle avoidance path, and the robot's safety cannot be guaranteed.

[0025] To address this, this application provides a method for dynamic obstacle avoidance in robots. The method involves acquiring raw environmental data of the target scene where the robot is located within a preset historical time period; obtaining historical trajectory data of a dynamic target within the target scene within the preset historical time period based on the raw environmental data; predicting the action intent of the dynamic target within the preset historical time period using a preset intent prediction network based on the historical trajectory data of the dynamic target within the preset historical time period, thereby obtaining action intent prediction information of the dynamic target at a preset future time; generating action decision information of the robot at a preset future time using a preset game theory model based on the action intent prediction information of the dynamic target at the preset future time, the robot's current state, and the robot's preset planned path for the target scene; and generating target control commands for the robot at the preset future time based on the action decision information. This application, by predicting the action intent of a dynamic target in the target scene and generating corresponding action decision information through a preset game theory model based on the predicted action intent information of the dynamic target and the robot's planned path, makes the interaction between the robot and the dynamic target smoother and improves the safety and rationality of the generated target control commands.

[0026] To clearly describe the robot dynamic obstacle avoidance method provided in the embodiments of this application, the method will be described below with reference to several accompanying drawings. The robot in this application can be an autonomous mobile robot, an autonomous vehicle, or an intelligent driving device such as a drone. The embodiments of this application are not intended to limit this. Figure 1 This is a flowchart illustrating a robot dynamic obstacle avoidance method provided in an embodiment of this application, as shown below. Figure 1 As shown, the method includes: Step 101: Obtain the original environmental data of the target scene where the robot is located during a preset historical time period.

[0027] The target scene can be an indoor space where humans and robots coexist, an urban scene, a semi-structured work area, a mixed traffic area, etc., and this application embodiment does not impose any limitations on this. The preset historical time period can be determined according to user needs, and can be 2 seconds, 3 seconds, etc., and this application embodiment does not impose any limitations on this. The original environmental data can include two-dimensional images, three-dimensional images, etc. of the target scene, and this application embodiment does not impose any limitations on this.

[0028] Optionally, by setting up corresponding data acquisition devices on the robot, the robot can continuously collect data to obtain the original environmental data of the target scene where the robot is located within a preset historical time period.

[0029] Step 102: Based on the original environmental data of the preset historical time period, obtain the historical trajectory data of the dynamic target in the target scene during the preset historical time period.

[0030] The historical trajectory data includes the position and velocity information of the dynamic target. The position information includes the position of the dynamic target in the horizontal direction and the position in the vertical direction, and the velocity information includes the velocity of the dynamic target in the horizontal direction and the velocity in the vertical direction.

[0031] Optionally, dynamic targets in the target scene are extracted based on the original environmental data of the preset historical time period, and historical trajectory data of the dynamic targets in the target scene during the preset historical time period are obtained based on the data corresponding to the dynamic targets in the original environmental data of the preset historical time period.

[0032] Step 103: Based on the historical trajectory data of the dynamic target in the preset historical time period, use the preset intention prediction network to predict the action intention of the dynamic target in the preset future time.

[0033] The preset intent prediction network is a pre-trained network used for intent prediction. The action intent prediction information includes: the dynamic action that the target may dynamically perform at each time point within a preset future time period, and the probability of performing that action. For example, the action intent prediction information at a given time point could be "70% straight ahead and 30% left turn". The time points within the preset future time period are the same as the original environmental data collected within a preset historical time period. For example, if the original environmental data for the preset historical time period consists of 20 pieces of original environmental data collected by the robot within 2 seconds, then the action intent prediction information for the preset future time period would be the prediction information of these 20 intention actions within those 2 seconds.

[0034] Optionally, based on the historical trajectory data of the dynamic target in a preset historical time period, a preset intention prediction network is used to predict the action intention of the dynamic target in a preset future time, wherein the preset future time is the same as the preset historical time period.

[0035] Step 104: Based on the predicted action intention of the dynamic target at a preset future time, the robot's current state, and the robot's preset planned path for the target scenario, a preset game model for the target scenario is used to generate the robot's action decision information at a preset future time.

[0036] Wherein, the robot's current state is the robot's current state vector. ,in, This represents the horizontal displacement. This represents the vertical displacement. The velocity is in the horizontal direction. The vertical velocity. The preset planned path is a path pre-defined by the robot within the target scene. The preset future time action decision information consists of multiple state vectors of the robot within the preset future time.

[0037] Optionally, based on the predicted action intention of the dynamic target at a preset future time, the robot's current state, and the robot's preset planned path for the target scenario, a preset game model for the target scenario is used to optimize the preset planned path, and the robot's action decision information at a preset future time is generated based on the optimized preset planned path.

[0038] Step 105: Based on the action decision information, generate target control instructions for the robot at a preset future time.

[0039] The target control commands are linear acceleration and angular acceleration.

[0040] Optionally, based on the action decision information, a target control command for the robot at a preset future time is generated, thereby controlling the corresponding robot according to the target control command.

[0041] In this embodiment, the original environmental data of the target scene where the robot is located within a preset historical time period is obtained; based on the original environmental data of the preset historical time period, historical trajectory data of the dynamic target in the target scene within the preset historical time period is obtained; based on the historical trajectory data of the dynamic target within the preset historical time period, a preset intention prediction network is used to predict the action intention of the dynamic target in a preset future time, thereby obtaining the action intention prediction information of the dynamic target in a preset future time; based on the action intention prediction information of the dynamic target in a preset future time, the current state of the robot, and the preset planned path of the robot for the target scene, a preset game model of the robot for the target scene is used to generate the action decision information of the robot in a preset future time; based on the action decision information, the target control command of the robot in a preset future time is generated. This application improves the adaptability and interactive intelligence of the robot in complex dynamic scenes by integrating dynamic target intention prediction driven by historical environmental data with action decision mechanism based on game model. It uses historical trajectory data and deep learning network to proactively predict the future behavior intentions of other dynamic targets, and combines the robot's own state and planned path to collaboratively optimize its own action decision within the game framework, achieving safer, more efficient, and more predictable motion control. This application avoids local optimum traps and frequent emergency braking problems, and improves the robustness of path generation.

[0042] Based on the above embodiments, the original environmental data includes point cloud data and image data. This application also provides a process for determining historical trajectory data in a robot dynamic obstacle avoidance method. Figure 2 This is a flowchart illustrating the process of determining historical trajectory data in a robot dynamic obstacle avoidance method provided in this application, as shown below. Figure 2 As shown, in step 102 above, based on the original environmental data of a preset historical time period, the historical trajectory data of the dynamic target in the target scene during the preset historical time period is obtained, including: Step 201: Perform semantic recognition on the image data and point cloud data to obtain the semantic labels of each point cloud cluster in the point cloud data.

[0043] The image data includes two-dimensional image data and three-dimensional image data. The two-dimensional image data is acquired by a depth camera installed on the robot. The two-dimensional images are color images, and the three-dimensional images are depth images. The point cloud data is acquired by a lidar installed on the robot.

[0044] Optionally, color images, depth images, and point cloud data are fused using a pre-calibrated extrinsic matrix on a depth camera to obtain fused point cloud data. Point cloud clusters are then determined based on the fused point cloud data, thereby obtaining semantic labels for each point cloud cluster. The fused point cloud data includes the semantic label for each point cloud.

[0045] Step 202: Based on the semantic tags of each point cloud cluster, obtain the historical trajectory data of the dynamic target in the preset historical time period from the point cloud data.

[0046] Among them, a point cloud cluster is a cluster of point clouds with the same semantic label, and each point cloud cluster corresponds to a dynamic target.

[0047] Optionally, based on the semantic labels of each point cloud cluster, the state information of the dynamic target in each point cloud data is obtained from the point cloud data of a preset historical time period, and the historical trajectory data of the dynamic target in the preset historical time period is determined by Kalman filtering based on the state information of the dynamic target in each point cloud data.

[0048] Optionally, the predicted state at the current moment can be obtained by using Kalman filtering based on the state of the dynamic target in each point cloud data at the previous moment.

[0049]

[0050]

[0051] in, Here is the state transition matrix. Sampling time, The process noise covariance matrix is... Let be the state covariance matrix. Predict the trajectory for the current moment. The trajectory from the previous moment. For the prior state covariance, Let be the posterior state covariance.

[0052] Optionally, based on the predicted state of the dynamic target at the current moment and the actual state of the point cloud data, the historical trajectory data of the dynamic target in a preset historical time period can be determined.

[0053]

[0054]

[0055]

[0056]

[0057]

[0058] in, for The actual state of point cloud data at any given time. for The difference between the actual state of the point cloud data at any given time and the predicted state at the current time. For the observation matrix, for Constantly innovate covariance To observe the noise covariance matrix, for Moment-time Kalman gain, for Constantly innovate covariance for Verify the state estimate after time step.

[0059] In this embodiment, image data and point cloud data are fused and semantically recognized to obtain semantic labels for each point cloud cluster. Based on the semantic labels of each point cloud cluster, historical trajectory data of a dynamic target within a preset historical time period is obtained from the point cloud data. This application, by fusing semantic recognition of image data and point cloud data, fully utilizes the rich texture and color information of images and the precise spatial geometric structure of point clouds to achieve a more accurate and robust semantic understanding of environmental objects. It accurately distinguishes dynamic targets from static backgrounds based on the semantic labels of point cloud clusters, thereby improving the accuracy and reliability of extracting historical trajectories of dynamic targets from complex point cloud data. This avoids trajectory discontinuity or deviation caused by misidentification or omission, providing a high-quality data foundation for subsequent intention prediction and motion decision-making, and significantly enhancing the robot's cognitive and responsive capabilities in dynamic open environments.

[0060] Based on the above embodiments, the historical trajectory data includes: the position and velocity of the dynamic target at multiple historical trajectory points. Therefore, this application also provides a process for determining action intention prediction information in a robot dynamic obstacle avoidance method. In step 103 above, based on the motion data of the dynamic target during a preset historical time period, a preset intention prediction network is used to predict the action intention of the dynamic target at a preset future time, including: Based on the position and velocity of the dynamic target at multiple historical trajectory points, an intention prediction network is used to predict the action intention of the dynamic target at a preset future time.

[0061] The action intent prediction information includes the probability of the dynamic target needle's intent to perform multiple preset actions at multiple time points within a preset future time. The preset actions can be actions that the robot can perform, such as moving straight, turning left, turning right, stopping, or crossing, and this embodiment does not impose any limitations on this. The intent prediction network is a Long Short-Term Memory (LSTM) intent prediction network. The LSTM intent prediction network includes: an input gate, a forget gate, candidate cell states, cell state updates, an output gate, and hidden state outputs.

[0062] Optionally, a historical trajectory sequence is generated based on the position and velocity of the dynamic target at multiple historical trajectory points. , to sequence historical trajectories The input gate, forget gate, candidate cell state, and output gate of the LSTM network are used to predict the corresponding first features. Second feature Third feature and the fourth feature , for the first feature Second feature and the third feature Perform cell state updates to obtain the fifth characteristic. According to the fourth feature and the fifth feature Hidden state features are obtained through hidden state data. Based on hidden state features Action intent prediction information is obtained through a softmax fully connected layer. .in, It is the sigmoid activation function. This indicates element-wise multiplication. and These are the trainable network weights and bias parameters.

[0063] In this embodiment, an intention prediction network is used to predict the action intention of a dynamic target at multiple historical trajectory points based on the target's position and velocity, thus obtaining the predicted action intention information of the dynamic target at a preset future time. By combining the position and velocity information of the dynamic target at multiple historical trajectory points for action intention prediction, this application preserves the spatiotemporal evolution characteristics of the target's motion, enabling the intention prediction network to more accurately capture its motion trends, acceleration / deceleration behaviors, and potential turning intentions. This improves the sensitivity and accuracy of judging future behavior, thereby enhancing the robot's ability to predict risks and make collaborative decisions in complex interactive scenarios.

[0064] Based on the above embodiments, this application also provides a process for determining a preset game model in a robot dynamic obstacle avoidance method. Figure 3 This application provides a flowchart illustrating the process of determining a preset game model for a robot dynamic obstacle avoidance method, as shown in the embodiments below. Figure 3 As shown, before generating the robot's action decision information at a preset future time in step 104 above, based on the predicted action intention information of the dynamic target at a preset future time, the robot's current state, and the robot's preset planned path for the target scenario, and using a preset game model for the target scenario, the method further includes: Step 301: Obtain the robot's first state set and the dynamic target's second state set.

[0065] The robot's preset state set includes: multiple reachable positions of the robot in the target scene and a preset speed range of the robot in the target scene. The dynamic target's second state set includes: multiple reachable positions of the dynamic target in the target scene and a preset speed range of the dynamic target in the target scene.

[0066] Optionally, in addition to the robot and the dynamic target, there are other obstacles in the target scene. Therefore, based on the target scene, we determine multiple locations other than the robot, the dynamic target, and other obstacles as multiple reachable locations for the robot in the target scene. Based on the speed limits of the target scene, we determine the robot's speed range in the target scene. Since the function of the dynamic target may differ from that of the robot, the reachable locations of the dynamic target may differ from those of the robot. Therefore, it is also necessary to determine the reachable locations of the dynamic target in the target scene. Similarly, the reachable locations of the dynamic target may differ from the robot's speed; therefore, it is also necessary to determine a preset speed range for the dynamic target in the target scene.

[0067] Step 302: Obtain the robot's first set of actions and the dynamic target's second set of actions.

[0068] The robot's first action set includes the actions that the robot can perform in the target scene, while the dynamic target's second action set includes the actions that the dynamic target can perform in the target scene.

[0069] Optionally, since the structures of the robot and the dynamic target may differ, the action set of the target robot and the action set of the dynamic target may be different. Therefore, a first action set of the robot is constructed based on the actions that the robot can perform, and a second action set is constructed based on the actions that the dynamic target can perform.

[0070] Step 303: Based on the first state set, the second state set, the first action set, and the second action set, construct a partially observable Markov game process model as the preset game model.

[0071] Among them, the Partially Observable Markov Game Process (POMGP) model is used by the robot to make predictions based on currently observable data.

[0072] Optionally, a partially observable Markov game process model is constructed based on the first state set, the second state set, the first action set, and the second action set as a preset game model. The reward function is a pre-trained function using multiple training data sets. The reward function includes: a safety reward function, an efficiency reward function, a comfort reward function, and an interaction reward function. The safety reward is a large negative reward given when the predicted distance to the pedestrian is less than a safety threshold; the efficiency reward encourages the robot to move towards the target point, with a larger positive reward for faster movement; the comfort reward penalizes severe linear and angular acceleration, which are the contents of the target control command; and the interaction reward is a positive reward given when a dynamic target object yields to the robot.

[0073] In this embodiment, a first state set of the robot and a second state set of the dynamic target are obtained, as well as a first action set of the robot and a second action set of the dynamic target. Based on the first state set, the second state set, the first action set, and the second action set, a partially observable Markov game process model is constructed as a preset game model. This application can fully characterize the interactive decision-making process between the robot and other intelligent agents in the environment under conditions of limited perception and incomplete information, enabling the robot to reasonably predict the behavior of the dynamic target in an uncertain environment and generate robust, safe, and interactively conscious action decisions.

[0074] Based on the above embodiments, this application also provides a process for determining action decision information in a robot obstacle avoidance method. Figure 4This is a flowchart illustrating the process of determining action decision information in an obstacle avoidance method for a robot provided in an embodiment of this application, as shown below. Figure 4 As shown, in step 103 above, based on the predicted action intention information of the dynamic target at a preset future time, the robot's current state, and the robot's preset planned path for the target scenario, a preset game model for the robot in the target scenario is used to generate the robot's action decision information at a preset future time, including: Step 401: Based on the historical trajectory data of the dynamic target in a preset historical time period and the predicted action intention information in a preset future time period, construct the observation set of the robot for the dynamic target.

[0075] Step 402: Based on the robot's current state, the preset planned path, and the observation set, a preset game model is used to solve the problem and generate the robot's action decision information at a preset future time.

[0076] The preset game model includes a corresponding solution strategy, which is the Monte Carlo tree search algorithm. The action decision information consists of the robot's position and velocity information at multiple points in a preset future time period.

[0077] Optionally, based on the robot's current state, the preset planned path, and the observation set, a solution strategy in a preset game model is used to solve the problem and generate the robot's action decision information for a preset future time.

[0078] In this embodiment, an observation set for the robot is constructed based on the historical trajectory data of the dynamic target within a preset historical time period and the predicted action intention information for a preset future time. A preset game theory model is then used to solve the observation set based on the robot's current state, the preset planned path, and the observation set, generating action decision information for the robot at the preset future time. This application constructs an observation set by combining the historical trajectory data of the dynamic target with the predicted action intention information, and uses this as input to incorporate the robot's current state and the preset planned path. By jointly solving the preset game theory model, it achieves a deep understanding and proactive response to the behavior of dynamic targets in the environment. This enables the robot to proactively predict conflict risks and identify interaction intentions in uncertain environments, and to generate safer, more efficient, and socially compliant action decisions in multi-entity competition or collaborative scenarios, thus improving the intelligence of autonomous navigation in complex dynamic environments.

[0079] Based on the above embodiments, this application also provides a process for generating target control commands in a robot dynamic obstacle avoidance method. Figure 5 This is a flowchart illustrating the generation of target control commands in a robot dynamic obstacle avoidance method provided in an embodiment of this application, as shown below. Figure 5As shown, in step 105 above, based on the action decision information, target control instructions for the robot at a preset future time are generated, including: Step 501: Generate a short-term trajectory for a preset future time based on the action decision information.

[0080] Optionally, based on the robot's position and speed information in the action decision information, a short-term trajectory of the robot at a preset future time can be generated.

[0081] For example, if the action decision information advocates for a gentle right-of-way, a smooth trajectory with a slight increase in linear velocity and a basically unchanged direction is generated. If the action decision information advocates for courteous yielding, a trajectory with a linear velocity reduced to 0 and a lateral offset to the right by a preset distance is generated. The preset distance can be 0.3 meters.

[0082] Step 502: Based on the current state, the preset planned path, and the short-term trajectory, the robot's preset dynamic constraint algorithm is used to solve the problem and obtain the target control command.

[0083] Among them, the preset dynamic constraint algorithm is to determine the state of the target robot at the next moment based on the current control command and the current state in the target control command.

[0084] In this embodiment, a short-term trajectory with a preset future time is generated based on action decision information. The target control command is obtained by solving a preset dynamic constraint algorithm for the robot based on the current state, the preset planned path, and the short-term trajectory. This application generates a short-term trajectory with a preset future time based on action decision information, and then combines this trajectory with the robot's current state, the preset planned path, and the short-term trajectory to generate the target control command using a preset dynamic constraint algorithm. This achieves a smooth, safe, and physically feasible closed-loop connection from high-level behavioral decision-making to low-level executable control, avoiding discontinuous, oscillating, or excessive control outputs, and improving the stability, safety, and execution accuracy of the motion process.

[0085] Based on the above embodiments, this application also provides a process for information transmission in a robot dynamic obstacle avoidance method. In addition to steps 101-105 above, the method further includes: If the dynamic target is a target agent, the short-term trajectory is communicated to the dynamic target.

[0086] In this context, the target intelligent agent is the same as the robot and can be controlled by a control device. Similarly, the target intelligent agent can also generate a corresponding short-term trajectory using the method described in this application.

[0087] Optionally, if the dynamic target is a target intelligent agent, the short-term trajectory is communicated to the dynamic target so that the dynamic target understands the robot's short-term trajectory, thereby enabling the interaction between multiple intelligent agents and the robot, and thus enabling subsequent game play.

[0088] In this embodiment, if the dynamic target is a target agent, a short-term trajectory is communicated to the dynamic target. This application enables bidirectional intent sharing and collaborative perception, improving the overall interactive transparency and collaboration efficiency of the multi-agent system. It allows each agent to make joint game decisions based on the prediction of the other's future movement, effectively reducing conflicts and conservative avoidance caused by misjudgment, thereby achieving more compact, safe, and efficient group behavior coordination.

[0089] Based on the above embodiments, this application also provides another process for determining target control commands in robot dynamic obstacle avoidance. Figure 6 This application provides another flowchart illustrating the determination of target control commands in dynamic obstacle avoidance for robots, as shown in the embodiments below. Figure 6 As shown, in step 502 above, based on the current state, the preset planned path, and the short-term trajectory, the robot's preset dynamic constraint algorithm is used to solve the problem, and the target control commands obtained include: Step 601: Based on the current state, obtain the robot's next state at the next moment.

[0090] The current state is The current state includes position and velocity information, as well as linear velocity and angular velocity. Let the robot's horizontal displacement be at the current moment. Let the vertical displacement of the robot at the current moment be denoted as . Let be the robot's angular velocity at the current moment. Let the robot's speed be at the current moment. Let be the linear velocity of the robot at the current moment.

[0091] Optionally, based on the current state, a preset calculation formula is used to obtain the robot's next state at the next moment.

[0092]

[0093]

[0094]

[0095]

[0096]

[0097]

[0098] in, For robots in Horizontal displacement at time t. For robots in Vertical displacement at time t, For robots in angular velocity at time t, For robots in The speed of time For robots in The linear velocity at time t. For robots in Horizontal displacement, For robots in The vertical displacement, For robots in angular velocity, For robots in speed, For robots in linear velocity, Sampling time, for linear acceleration at time t, for The angular acceleration at a given time, the target control command at the current time includes the linear acceleration and the angular acceleration at the current time. For robots in The maximum value of each parameter at time t.

[0099] Step 602: Based on the robot's next state and the preset state corresponding to the next moment on the preset planned path, obtain the first variable parameter.

[0100] Optionally, according to the robot in the first time and on the pre-planned path at the first Preset state corresponding to the time and the corresponding weighting coefficients To obtain the first variable parameter .

[0101] Step 603: Based on the robot's next state and the predicted state corresponding to the next moment on the short-term trajectory, obtain the second variable parameter.

[0102] Optionally, based on the robot's position in the first... time and in the short-term trajectory at the first Predicted state at time 1 and the corresponding weighting coefficients To obtain the second variable parameter .

[0103] Step 604: Based on the robot's next state and the pre-acquired next state of the dynamic target at the next moment, obtain the third variable parameter.

[0104] Among them, the next state of the dynamic target at the next moment, which is obtained in advance, can be the next state of the dynamic target at the next moment determined in the short-term trajectory sent by the dynamic target controller.

[0105] Optionally, the position information of the robot in the next state is determined based on the robot's next state, and the position information of the dynamic target in the next state is determined based on the pre-acquired next state of the dynamic target, thus determining the position information of the target in the next state at the next moment. At that moment, the robot and the first Euclidean distance between dynamic targets According to the robot and the first Euclidean distance between dynamic targets Parameters for controlling the range of repulsive force from obstacles and the corresponding weights To obtain the third variable parameter .

[0106] Step 605: Obtain the fourth variable parameter based on multiple preset control commands.

[0107] Optionally, according to the first Preset control commands for the time and the corresponding weights To obtain the fourth variable parameter .in, , This represents the maximum value of the parameter in the instruction.

[0108] Step 606: Construct the objective function based on the first variable parameter, the second variable parameter, the third variable parameter, and the fourth variable parameter.

[0109]

[0110] in, To pre-determine multiple time domains in the future, The minimum path is determined under the control of the control command U.

[0111] Step 607: Solve the preset dynamic constraint algorithm according to the objective function to determine the optimal target control command from multiple preset control commands.

[0112] The preset control commands include a set of multiple linear accelerations and angular accelerations, with different control commands corresponding to different objective function results.

[0113] Optionally, based on the objective function, a preset dynamic constraint algorithm is solved. By substituting different instructions from the preset control instructions into the objective function, the optimal target control instruction is determined from multiple preset control instructions as the target control instruction.

[0114] In this embodiment, by comprehensively considering the deviation between the robot's next state and the preset planned path, the consistency with the expected state of the short-term trajectory, the interaction with the future state of the dynamic target, and the smoothness and feasibility of the candidate control commands themselves, a multi-dimensional joint optimization objective function is constructed. Based on this, the optimal control command under dynamic constraints is solved, realizing multi-level collaborative optimization from environmental adaptation, path tracking, interaction safety to control feasibility. This not only enhances the robot's robust response capability to sudden disturbances in dynamic and complex scenarios, but also ensures that the selected target control command has higher safety, stability, and social rationality while satisfying physical constraints by quantifying and balancing the global planning intention, local prediction behavior, and multi-agent interaction risks, thereby improving the intelligence level and actual control accuracy of autonomous motion decision-making.

[0115] Based on the same inventive concept, this application also provides a robot dynamic obstacle avoidance device corresponding to the robot dynamic obstacle avoidance method. Since the principle of the device in this application is similar to the robot dynamic obstacle avoidance method described above in this application, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be described again.

[0116] Figure 7 This is a schematic diagram of the structure of a robot dynamic obstacle avoidance device provided in an embodiment of this application, as shown below. Figure 7 As shown, the device includes: The first acquisition module 701 is used to acquire the original environmental data of the target scene where the robot is located during a preset historical time period; The second acquisition module 702 is used to acquire historical trajectory data of dynamic targets in the target scene during the preset historical time period based on the original environmental data of the preset historical time period. Prediction module 703 is used to predict the action intent of the dynamic target in a preset future time by using a preset intent prediction network based on the historical trajectory data of the dynamic target in a preset historical time period. The first generation module 704 is used to generate the robot's action decision information at a preset future time based on the dynamic target's action intention prediction information at a preset future time, the robot's current state, and the robot's preset planned path for the target scene, using a preset game model for the target scene. The second prediction module 705 is used to generate target control instructions for the robot at a preset future time based on the action decision information.

[0117] In one possible implementation, the second acquisition module 702 is specifically used to: perform fusion semantic recognition on image data and point cloud data to obtain semantic labels for each point cloud cluster in the point cloud data; Based on the semantic labels of each point cloud cluster, historical trajectory data of dynamic targets within a preset historical time period are obtained from the point cloud data.

[0118] In one possible implementation, the historical trajectory data includes: the position and velocity of the dynamic target at multiple historical trajectory points; the prediction module 703 is specifically used to: based on the position and velocity of the dynamic target at multiple historical trajectory points, use an intention prediction network to predict the action intention of the dynamic target at a preset future time, and obtain the action intention prediction information of the dynamic target at a preset future time, the action intention prediction information including: the probability of the dynamic target's intention to multiple preset actions.

[0119] In one possible implementation, the first generation module 703 is further configured to: acquire a first state set of the robot and a second state set of the dynamic target, wherein the preset state set of the robot includes: multiple reachable positions of the robot in the target scene and a preset speed range of the robot in the target scene, and the second state set of the dynamic target includes: multiple reachable positions of the dynamic target in the target scene and a preset speed range of the dynamic target in the target scene. Obtain a first set of actions for the robot and a second set of actions for the dynamic target, wherein the first set of actions for the robot includes actions that the robot can perform in the target scene, and the second set of actions for the dynamic target includes actions that the dynamic target can perform in the target scene; Based on the first set of states, the second set of states, the first set of actions, and the second set of actions, a partially observable Markov game process model is constructed as the preset game model.

[0120] In one possible implementation, the first generation module 704 is specifically used to: construct an observation set for the robot targeting the dynamic target based on the historical trajectory data of the dynamic target in a preset historical time period and the action intention prediction information of a preset future time. Based on the robot's current state, the pre-planned path, and the observation set, a pre-defined game theory model is used to solve the problem and generate the robot's action decision information for a pre-defined future time.

[0121] In one possible implementation, the second generation module 705 is specifically used to: generate a short-term trajectory for a preset future time based on action decision information; Based on the current state, the preset planned path, and the short-term trajectory, the robot's preset dynamic constraint algorithm is used to solve the problem and obtain the target control command.

[0122] In one possible implementation, the device further includes a transmitting module, specifically configured to: if the dynamic target is a target intelligent agent, transmit a short-term trajectory to the dynamic target.

[0123] In one possible implementation, the second generation module 705 is specifically used to: obtain the robot's next state at the next moment based on the current state; Based on the robot's next state and the preset state corresponding to the next moment on the preset planned path, the first variable parameter is obtained; The second variable parameter is obtained based on the robot's next state and the predicted state at the next moment on the short-term trajectory; The third variable parameter is obtained based on the robot's next state and the pre-acquired next state of the dynamic target at the next moment; The fourth variable parameter is obtained based on multiple preset control commands; Construct the objective function based on the first, second, third, and fourth variable parameters; Based on the objective function, a preset dynamic constraint algorithm is solved to determine the optimal target control command from multiple preset control commands.

[0124] The processing flow of each module in the device and the interaction flow between each module can be referred to the relevant descriptions in the above method embodiments, and will not be detailed here.

[0125] This application also provides a control device. Figure 8 A schematic diagram of a control device structure is provided for an embodiment of this application, such as... Figure 8 As shown, the control device includes a processor 801 and a memory 802, and optionally, a bus 803. The memory 802 stores machine-readable instructions executable by the processor 801. When the control device 800 is running, the processor 801 and the memory 802 communicate via the bus 803. When the machine-readable instructions are executed by the processor 801, the steps of the above-described robot dynamic obstacle avoidance method are performed.

[0126] This application also provides a robot. Figure 9 A robot provided in the embodiments of this application, such as Figure 9 As shown, the robot includes a robot body 901 and a control device 800 disposed within the robot body. The control device is used to execute the steps of the above-described robot dynamic obstacle avoidance method.

[0127] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the above-described robot dynamic obstacle avoidance method.

[0128] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and devices described above can be referred to the corresponding processes in the method embodiments, and will not be repeated here. In the several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection can be through some communication interfaces; the indirect coupling or communication connection of devices or modules can be electrical, mechanical, or other forms.

[0129] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. If the functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a control device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0130] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A dynamic obstacle avoidance method for robots, characterized in that, The method includes: Obtain raw environmental data of the target scene where the robot is located within a preset historical time period; Based on the original environmental data of the preset historical time period, obtain the historical trajectory data of the dynamic target in the target scene during the preset historical time period; Based on the historical trajectory data of the dynamic target in the preset historical time period, a preset intention prediction network is used to predict the action intention of the dynamic target in the preset future time to obtain the action intention prediction information of the dynamic target in the preset future time. Based on the predicted action intention information of the dynamic target at the preset future time, the current state of the robot, and the preset planned path of the robot for the target scenario, the robot's action decision information at the preset future time is generated using the preset game model of the robot for the target scenario. Based on the action decision information, the robot generates target control instructions for the preset future time.

2. The method according to claim 1, characterized in that, The original environmental data includes point cloud data and image data; obtaining the historical trajectory data of the dynamic target in the target scene within the preset historical time period based on the original environmental data of the preset historical time period includes: The image data and the point cloud data are fused and semantically recognized to obtain the semantic labels of each point cloud cluster in the point cloud data; Based on the semantic tags of each point cloud cluster, the historical trajectory data of the dynamic target during the preset historical time period is obtained from the point cloud data.

3. The method according to claim 1, characterized in that, The historical trajectory data includes: the position and velocity of the dynamic target at multiple historical trajectory points; The step of predicting the motion intent of the dynamic target in a preset future time by using a preset intent prediction network based on the motion data of the dynamic target in the preset historical time period includes: Based on the position and velocity of the dynamic target at multiple historical trajectory points, the intention prediction network is used to predict the action intention, thereby obtaining the action intention prediction information of the dynamic target at a preset future time. The action intention prediction information includes the probability of the dynamic target's intention to multiple preset actions.

4. The method according to claim 1, characterized in that, Before generating the robot's action decision information at the preset future time based on the predicted action intention information of the dynamic target at the preset future time, the robot's current state, and the robot's preset planned path for the target scenario, and using a preset game model for the target scenario, the method further includes: A first state set of the robot and a second state set of the dynamic target are obtained, wherein the preset state set of the robot includes: multiple reachable positions of the robot in the target scene and a preset speed range of the robot in the target scene, and the second state set of the dynamic target includes: multiple reachable positions of the dynamic target in the target scene and a preset speed range of the dynamic target in the target scene; Obtain a first set of actions for the robot and a second set of actions for the dynamic target, wherein the first set of actions for the robot includes actions that the robot can perform in the target scene, and the second set of actions for the dynamic target includes actions that the dynamic target can perform in the target scene; Based on the first state set, the second state set, the first action set, and the second action set, a partially observable Markov game process model is constructed as the preset game model.

5. The method according to claim 1, characterized in that, The step of generating action decision information for the robot at the preset future time based on the predicted action intention information of the dynamic target at the preset future time, the current state of the robot, and the preset planned path of the robot for the target scenario, using a preset game model of the robot for the target scenario, includes: Based on the historical trajectory data of the dynamic target during the preset historical time period and the action intention prediction information for the preset future time, the robot constructs an observation set for the dynamic target. Based on the robot's current state, the preset planned path, and the observation set, the preset game model is used to solve the problem and generate the robot's action decision information at the preset future time.

6. The method according to claim 1, characterized in that, The step of generating the target control command for the robot at the preset future time based on the action decision information includes: Based on the action decision information, a short-term trajectory for the preset future time is generated; Based on the current state, the preset planned path, and the short-term trajectory, the target control command is obtained by solving the preset dynamic constraint algorithm of the robot.

7. The method according to claim 5, characterized in that, The step of solving the target control command by using the robot's preset dynamic constraint algorithm based on the current state, the preset planned path, and the short-term trajectory includes: Based on the current state, obtain the next state of the robot at the next moment; Based on the robot's next state and the preset state corresponding to the next moment on the preset planning path, the first variable parameter is obtained; The second variable parameter is obtained based on the robot's next state and the predicted state corresponding to the next moment on the short-term trajectory; The third variable parameter is obtained based on the robot's next state and the pre-acquired next state of the dynamic target at the next moment; The fourth variable parameter is obtained based on multiple preset control commands; Based on the first variable parameter, the second variable parameter, the third variable parameter, and the fourth variable parameter, construct the objective function for solving; Based on the objective function, the preset dynamic constraint algorithm is solved to determine the optimal target control command from the plurality of preset control commands.

8. A dynamic obstacle avoidance device for robots, characterized in that, The device includes: The first acquisition module is used to acquire the original environmental data of the target scene where the robot is located during a preset historical time period; The second acquisition module is used to acquire historical trajectory data of dynamic targets in the target scene during the preset historical time period based on the original environmental data of the preset historical time period. The prediction module is used to predict the action intent of the dynamic target in a preset future time by using a preset intent prediction network based on the historical trajectory data of the dynamic target in the preset historical time period. The first generation module is used to generate the robot's action decision information at the preset future time based on the predicted action intention information of the dynamic target at the preset future time, the current state of the robot, and the preset planned path of the robot for the target scenario, using a preset game model of the robot for the target scenario. The second generation module is used to generate target control instructions for the robot at the preset future time based on the action decision information.

9. A control device, characterized in that, include: The processor and memory, the memory storing machine-readable instructions executable by the processor, wherein when the control device is running, the processor executes the machine-readable instructions to perform the steps of the robot dynamic obstacle avoidance method as described in any one of claims 1 to 8.

10. A robot, characterized in that, The robot includes at least: a robot body and a control device disposed within the robot body, the control device being used to perform the steps of the robot dynamic obstacle avoidance method as described in any one of claims 1 to 8.

Images