Risk-adaptive safe navigation method and device, terminal and storage medium

Abstract

The invention provides a risk-adaptive safe navigation method and device, a terminal and a storage medium, and belongs to the technical field of robot control, and the method comprises the following steps: generating an environment state vector at a current moment; obtaining a risk assessment value; adjusting a security constraint area of the control barrier function and weight distribution of a risk weighted cost function in the model prediction control module; performing optimization solution based on the adjusted risk weighted cost function and the safety constraint formed by the adjusted safety constraint region by using a model prediction control module to obtain a target control instruction meeting the safety constraint and the optimal cost; the robot is driven to move. According to the method, the environment risk is dynamically modeled, the weights of the safety constraint condition and the risk cost function are adaptively adjusted, the adjusted safety constraint condition and the risk cost function are introduced into the model prediction control module to generate the target control instruction, the method can dynamically adapt to the change of the environment, and then the accuracy of navigation is ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of robot control, and particularly relates to a risk adaptive safe navigation method and device, a terminal and a storage medium. BACKGROUND

[0002] With the wide application of intelligent mobile robots in service industry, industrial production, warehouse logistics and social public places, it has become a research hotspot to realize safe, smooth and efficient autonomous navigation of robots in dynamic uncertain environments. In the prior art, based on traditional optimization theory and model predictive control (MPC), it is the mainstream research direction. This kind of method solves the optimization problem in the prediction time domain to generate control input by explicitly modeling the system dynamics and environmental constraints. Its method includes Nominal MPC, Tube-based MPC and Robust MPC. Among them, Tube-based MPC constructs a robust tube around the predicted trajectory to ensure safety under the action of disturbance and uncertainty; Robust MPC regards environmental uncertainty as a bounded disturbance and optimizes for the worst case to ensure the stability of system performance.

[0003] Although the above method provides strong stability and constraint feasibility in theory, it is highly dependent on the accuracy of the dynamic model and the certainty of the environmental parameters. When the model has deviation or the environment changes dramatically, the control effect will decrease significantly, and the accuracy of navigation cannot be guaranteed.

[0004] Therefore, the prior art has defects and needs to be improved and developed. SUMMARY

[0005] The technical problem solved by the present application is to provide a risk adaptive safe navigation method, device, terminal and storage medium to solve the problem that the prior art cannot guarantee the accuracy of navigation.

[0006] The technical solution adopted by the present application to solve the technical problem is as follows: In a first aspect, the present application provides a risk adaptive safe navigation method, which comprises: acquiring and generating an environmental state vector at the current time based on the current state information of the robot, the motion information of the obstacle and the environmental information at a preset frequency; inputting the environmental state vector at the current time into a learning module for processing to obtain a risk assessment value; adjusting the safety constraint region of the control obstacle function and the weight distribution of the risk weighted cost function in the model predictive control module based on the risk assessment value; The model predictive control module is used to perform optimization and solving based on the adjusted risk weighted cost function and the safety constraint composed of the adjusted safety constraint region, to obtain a target control instruction meeting the safety constraint and optimal cost; The target control instruction is sent to an execution mechanism of the robot to drive the robot to move.

[0007] In an embodiment, after the target control instruction is sent to the execution mechanism of the robot to drive the robot to move, the method further comprises: Collecting trajectory deviation, risk change data and safety constraint state after the robot moves as feedback data; The feedback data is fed back to the learning module and the model predictive control module to update the policy parameter of the learning module and the control strategy of the model predictive control module.

[0008] In an embodiment, based on the risk evaluation value, the safety constraint region of the control barrier function is adjusted, comprising: If the risk evaluation value is higher than a preset first threshold value, the safety constraint region is reduced; If the risk evaluation value is lower than a preset second threshold value, the safety constraint region is increased.

[0009] In an embodiment, based on the risk evaluation value, the weight distribution of the risk weighted cost function in the model predictive control module is adjusted, comprising: If the risk evaluation value is higher than a preset first threshold value, the weight of the safety performance index in the risk weighted cost function in the model predictive control module is adjusted to be higher; If the risk evaluation value is lower than a preset second threshold value, the weight of the motion efficiency index in the risk weighted cost function in the model predictive control module is adjusted to be higher.

[0010] In an embodiment, the learning module comprises a policy network trained based on a proximal policy optimization algorithm.

[0011] In an embodiment, the process of optimization and solving by the model predictive control module comprises: The model predictive control module is used to generate a predicted state corresponding to a multi-step future trajectory in a preset prediction time domain; At each time in the prediction time domain, the predicted state is verified whether it meets the safety constraint composed of the adjusted safety constraint region; The predicted trajectory branch meeting the safety constraint at all times is reserved, and an optimization calculation is performed in combination with the adjusted risk weighted cost function, to obtain a target space instruction meeting the safety constraint and optimal cost.

[0012] In an embodiment, the method further comprises: The execution of the iteration generates an environment state vector, obtains a risk assessment value, generates a target control instruction, executes control, collects feedback data, and updates until the navigation task is completed.

[0013] In a second aspect, the embodiments of the present application further provide a risk adaptive safe navigation system, which comprises: A vector generation module is configured to obtain and generate an environment state vector at a preset frequency based on current state information of the robot, motion information of the obstacle, and environment information; A risk assessment module is configured to input the environment state vector at the current time into a learning module for processing to obtain a risk assessment value; An adjustment module is configured to adjust a safety constraint region of a control obstacle function and a weight distribution of a risk weighted cost function in a model predictive control module based on the risk assessment value; An instruction generation module is configured to use the model predictive control module to perform optimization and solving based on the adjusted risk weighted cost function and a safety constraint formed by the adjusted safety constraint region to obtain a target control instruction that satisfies the safety constraint and is optimal in cost; A driving module is configured to send the target control instruction to an execution mechanism of the robot to drive the robot to move.

[0014] In a third aspect, the embodiments of the present application further provide a terminal, which comprises a memory, a processor, and a risk adaptive safe navigation program stored in the memory and executable on the processor, and the risk adaptive safe navigation program, when executed by the processor, implements the steps of the risk adaptive safe navigation method as described above.

[0015] In a fourth aspect, the embodiments of the present application further provide a computer readable storage medium, which stores a risk adaptive safe navigation program, and the risk adaptive safe navigation program can be executed to implement the steps of the risk adaptive safe navigation method as described above.

[0016] The application has the beneficial effects that: the application generates the environment state vector of the current moment, obtains the risk evaluation value, adjusts the safety constraint region of the control barrier function and the weight distribution of the risk weighted cost function in the model predictive control module, uses the model predictive control module to perform optimization and solving based on the adjusted risk weighted cost function and the safety constraint formed by the adjusted safety constraint region, obtains the target control instruction meeting the safety constraint and the optimal cost, and drives the robot movement. The application dynamically models the environmental risk, adaptively adjusts the safety constraint condition and the risk cost function weight, introduces the adjusted safety constraint condition and the risk cost function into the model predictive control module to generate the target control instruction, can dynamically adapt to the change of the environment, and further ensures the accuracy of the navigation. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 is a flow chart of the preferred embodiment of the risk adaptive safety navigation method in the application.

[0018] Figure 2 is a schematic diagram of the learning module training and application.

[0019] Figure 3 is a structure schematic diagram of the preferred embodiment of the risk adaptive safety navigation system in the application.

[0020] Figure 4 is a terminal principle block diagram of the application. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical scheme and advantages of the application more clear and definite, the application is further described in detail below with reference to the drawings and examples. It should be understood that the specific examples described herein are only used to explain the application, and are not used to limit the application.

[0022] With the wide application of intelligent mobile robots in service industry, industrial production, warehouse logistics and social public places, it has become a research hotspot to realize safe, smooth and efficient autonomous navigation of robots in dynamic uncertain environments. Traditional navigation systems usually include three parts: global path planning, local obstacle avoidance control and trajectory tracking. With the increasing complexity of the scene, control methods relying only on static maps or heuristic rules cannot meet the safety and stability requirements in dynamic scenarios. To cope with complex dynamic environments (such as moving obstacles, non-deterministic moving bodies and crowd behavior), researchers began to introduce model predictive control (MPC) modules to balance path tracking errors, control costs and safety constraints within the prediction horizon through online optimization of control inputs. MPC can handle system dynamics and constraints to some extent and is a typical high-performance control strategy. However, in practical applications, there are often uncertainties in environmental perception and dynamic prediction: sensor noise, obstacle motion prediction errors, map defects and other factors can lead to risk accumulation in the planning and control process. In order to improve safety in uncertain environments, the research trend in recent years is to introduce risk assessment and uncertainty modeling into control optimization, so that the control strategy not only considers performance optimization, but also minimizes risk. In this context, research on learning methods and control barrier functions has gradually emerged. The former can learn the environment model and risk distribution through data-driven learning, and the latter can guarantee system safety through formal mathematical constraints. The goal of such methods is to achieve a navigation control mechanism that is "learning-guided, risk-adaptive and safety-certified".

[0023] In the prior art, methods based on traditional optimization theory and model predictive control (MPC) are the mainstream research direction. These methods include Nominal MPC, Tube-based MPC and Robust MPC. Among them, Tube-based MPC constructs a robust tube around the predicted trajectory to ensure safety under the action of disturbances and uncertainties; Robust MPC considers environmental uncertainties as bounded disturbances and optimizes for the worst case to ensure the stability of system performance.

[0024] Although the above methods provide strong stability and constraint feasibility in theory, they are highly dependent on the accuracy of the dynamic model and the certainty of the environmental parameters. When the model has deviations or the environment changes dramatically, the control effect will decrease significantly, and the accuracy of navigation cannot be guaranteed.

[0025] In view of the above defects of the prior art, the present application provides a risk adaptive safe navigation method and device, a terminal and a storage medium, the method comprising: generating an environment state vector at the current time; obtaining a risk evaluation value; adjusting the safety constraint region of the control barrier function and the weight distribution of the risk weighted cost function in the model predictive control module; using the model predictive control module to perform optimization and solving based on the adjusted risk weighted cost function and the safety constraint formed by the adjusted safety constraint region, to obtain a target control instruction that satisfies the safety constraint and is optimal in cost; and driving the robot to move. The present application dynamically models the environmental risk, adaptively adjusts the safety constraint condition and the risk cost function weight, and introduces the adjusted safety constraint condition and the risk cost function into the model predictive control module to generate the target control instruction, which can dynamically adapt to changes in the environment and thus ensure the accuracy of navigation.

[0026] Referring to Figure 1 , the risk adaptive safe navigation method described in the embodiments of the present application comprises the following steps: Step S100, generating an environment state vector based on the current state information of the robot, the motion information of the obstacles and the environmental information at a preset frequency.

[0027] Specifically, raw data is collected in real time by the sensor system (such as a laser radar, a camera, an inertial measurement unit, etc.) of the robot. These data mainly include the current state information of the robot, the motion information of the obstacles and the environmental information, to generate the environment state vector at the current time. The current state information of the robot includes the position coordinates, the heading angle, the linear velocity, the angular velocity, etc. of the robot. The motion information of the obstacles includes the position coordinates, the motion velocity, the motion direction, the shape information, etc. of the obstacles. The environmental information includes the width of the road, the range of the feasible region, the position coordinates of the terminal, etc. The environment state vector can be used as the input information of the subsequent learning module, to provide basic data for the subsequent risk evaluation and strategy optimization. In this way, real-time and structured environment state input can be provided for the control decision, to improve the perception accuracy and the decision response speed.

[0028] Referring to Figure 1 , the risk adaptive safe navigation method described in the embodiments of the present application further comprises the following steps: Step S200, inputting the environment state vector at the current time into a trained learning module for processing, to obtain a risk evaluation value.

[0029] Specifically, the trained learning module includes a strategy network trained based on a proximal policy optimization algorithm. The schematic diagram of the learning module training and application is shown in Figure 2 . In the training phase, the raw state at the current time is obtained in the simulation environment. (including current state information of the robot, motion information of the obstacle, and environment information, etc.), and converts it into an environment state vector through a feature extractor and inputs it into a policy network The policy network outputs a risk probability distribution After sampling the risk probability distribution, a selected risk assessment value is input into a CVaR-CBF module, which uses a preset risk threshold in combination with the risk assessment value to perform compliance verification on the action, and outputs a correction suggestion if the action violates the safety constraint, and finally outputs an executable action The simulator executes the action , the environment transitions to a new state , and generates a reward signal The complete experience of this step is stored in an experience replay buffer. Based on the proximal policy optimization algorithm, data is sampled from the buffer to update the parameters of the policy network stably and efficiently The goal is to maximize the long-term cumulative reward. Through iteration, the above steps are trained, and when the preset end condition is reached, the training is completed. After the training is completed, the trained policy network is fixed in the learning module, and when the environment state vector is received, the corresponding risk assessment value can be obtained.

[0030] Please refer to Figure 1 , the risk adaptive safety navigation method described in the embodiments of the present application further comprises the following steps: Step S300, based on the risk assessment value, adjust the safety constraint region of the control barrier function and the weight distribution of the risk weighted cost function in the model predictive control module.

[0031] Specifically, based on the risk assessment value, the safety constraint region of the control barrier function is adjusted, including: if the risk assessment value is higher than a preset first threshold, the safety constraint region is reduced. At this time, it is considered as a high-risk situation. The system will take more conservative measures to reduce the safety constraint region to ensure that the robot and the obstacle maintain a larger safety margin, thereby reducing the probability of collision. If the risk assessment value is lower than a preset second threshold, the safety constraint region is increased. At this time, it is considered as a low-risk situation. The system will take more aggressive measures, and increasing the safety constraint region means relaxing the safety constraint, allowing the robot to be closer to the obstacle, which provides a basis for improving navigation efficiency. The first threshold is higher than the second threshold. The present application dynamically shrinks or relaxes the safety constraint according to the environmental risk in the above manner, thereby strengthening the safety protection in the high-risk area and maintaining flexibility in the low-risk area, ensuring that the system state is always within the theoretical safety set. This design realizes risk-sensitive adjustment of the safety boundary, significantly improving the obstacle avoidance reliability and formal safety guarantee capability.

[0032] In addition, the weight allocation of the risk-weighted cost function in the model predictive control module will be adjusted based on the risk assessment value. Specifically, if the risk assessment value is higher than a preset first threshold, the weight of the safety performance index in the risk-weighted cost function of the model predictive control module will be increased. This is considered a high-risk situation, and the model predictive control module will prioritize safety when solving the problem. If the risk assessment value is lower than a preset second threshold, the weight of the motion efficiency index in the risk-weighted cost function of the model predictive control module will be increased. This is considered a low-risk situation, and the model predictive control module will prioritize motion efficiency when solving the problem, allowing it to focus more on planning faster and more energy-efficient trajectories. This approach enables the robot to dynamically adapt to uncertain environments and make decisions best suited to those environments. By introducing the risk assessment value into the model predictive control module and the control obstacle function, the controller can dynamically adjust the safety margin and control weights according to the risk level, achieving a strategy switch of "conservative in high-risk, efficient in low-risk," thereby endowing the system with the ability to recognize and adapt to environmental risks, effectively balancing safety and performance.

[0033] Please see Figure 1 The risk-adaptive safety navigation method described in this embodiment of the invention further includes the following steps: Step S400: Using the model predictive control module, optimize the solution based on the adjusted risk-weighted cost function and the safety constraints formed by the adjusted safety constraint region to obtain the target control command that satisfies the safety constraints and minimizes the cost.

[0034] Specifically, the optimization process using the model predictive control module includes: generating predicted states corresponding to multiple future trajectories within a preset prediction time domain; verifying whether the predicted states satisfy the safety constraints constituted by the adjusted safety constraint region at each moment within the prediction time domain; retaining the predicted trajectory branches that satisfy the safety constraints at all moments, and performing optimization calculations in conjunction with the adjusted risk-weighted cost function to obtain the target space instruction that satisfies the safety constraints. By performing mandatory safety verification on the predicted states at each moment within the prediction time domain, it is ensured that the generated trajectory, from the current state to the future state, strictly lies within the safe and feasible set defined by the control obstacle function. This provides a safety guarantee not available in traditional methods, fundamentally avoiding collision risks caused by single-step decision errors or environmental uncertainties. Dynamic risk assessment values ​​are deeply integrated into the optimization process: on the one hand, as hard constraints limiting the solution space; on the other hand, the risk-weighted cost function adjusts the optimization target weights. This dual adjustment mechanism allows the solver to dynamically adjust its optimization preferences according to the risk level, making the predicted trajectory more conservative in high-risk regions and smoother in low-risk regions.

[0035] Referring to Figure 1 The risk adaptive safe navigation method according to the embodiment of the application further comprises the following steps: Step S500: sending the target control instruction to an execution mechanism of the robot to drive the robot to move.

[0036] Specifically, the target control instruction can drive the robot to move according to the uncertain environment. Referring to Figure 2 After the trained learning module is deployed, the received environment state vector is processed to obtain a risk probability distribution output by the policy network The risk probability distribution is quantile to obtain a risk evaluation value Based on the risk evaluation value, the safety constraint region of the control barrier function and the weight distribution of the risk weighted cost function in the model predictive control module are adjusted, and the model predictive control module is used to perform optimization and solving based on the adjusted risk weighted cost function and the safety constraint formed by the adjusted safety constraint region, to obtain a target control instruction that meets the safety constraint and has the minimum cost, and then the target control instruction is used to drive the robot to move.

[0037] In an implementation manner, after the target control instruction is sent to the execution mechanism of the robot to drive the robot to move, the method further comprises the following steps: Collecting trajectory deviation, risk change data and safety constraint state after the robot moves as feedback data; The feedback data is returned to the learning module and the model predictive control module to update the policy parameters of the learning module and the control strategy of the model predictive control module.

[0038] Specifically, the parameters are updated by returning the feedback data, which can realize continuous improvement of the policy and dynamic correction of the risk.

[0039] In an implementation manner, the method further comprises the following steps: The steps of generating an environment state vector, obtaining a risk evaluation value, generating a target control instruction, performing control, collecting feedback data and updating are iteratively executed until a navigation task is completed.

[0040] Specifically, in the iterative process, the policy network continuously updates the policy parameters and the control strategy of the model predictive control module according to the feedback, forming a self-learning and adaptive control system, and effectively realizing safe and interpretable autonomous control of the robot.

[0041] In the prior art, in addition to the problem that the control effect will be significantly reduced when the model has deviation or the environment changes drastically, the control strategy is often too conservative due to the need to consider the worst-case disturbance in the prediction process, resulting in reduced navigation efficiency and difficulty in balancing safety and performance. In addition, the learning-enhanced control method (Learning-Augmented Control) and the safety control method based on control barrier function (CBF) in the prior art are also the main research direction. The learning-enhanced control method (Learning-Augmented Control) mainly uses machine learning technology to enable the controller to have self-learning and prediction ability. For example: Learning-based MPC (LMPC) can use historical trajectory data to learn the cost function or dynamic model; Gaussian process MPC (GP-MPC) captures system nonlinearity and environmental uncertainty through non-parametric modeling, thereby improving prediction accuracy. Some research also uses deep neural network (DNN) or reinforcement learning (RL) framework to directly learn the obstacle behavior model or control strategy. Although the learning method improves the environmental understanding and modeling ability of the system, its safety and stability still lack theoretical guarantee. The "black box" nature of neural network makes it less interpretable, and when encountering scenes outside the training distribution, performance degradation or even safety failure may occur. In addition, the learning model is highly dependent on the quality of training data, and if the data coverage is not complete, it may lead to unstable navigation strategy or unpredictable behavior. Control barrier function (Control Barrier Function, CBF) is a formal safety constraint tool that can ensure that the system state is always within the safe set (Safe Set). Control barrier function can be combined with model predictive control (MPC) to form a safe-MPC optimization framework, which prevents the system from exceeding the boundary or colliding by introducing safety function constraints in the optimization process. Some research proposes adaptive control barrier function and robust control barrier function based on this to deal with disturbances and modeling errors. Although control barrier function can provide safety guarantee in theory, its performance is strongly dependent on the definition of safety set and parameter selection. When the obstacle motion has high uncertainty or the prediction error is large, the control barrier function constraint may fail, resulting in incomplete safety guarantee. At the same time, the traditional control barrier function is usually designed for deterministic systems, and lacks adaptive ability to risk distribution and random uncertainty, thereby limiting its application in dynamic complex environment. The present invention realizes risk perception and adaptive enhancement, and the controller can automatically adjust the control strategy according to the environmental risk; by introducing control barrier function constraint to ensure that the system state is always within the safe feasible range in theory, it has formal safety guarantee. It can effectively overcome the shortcomings of the above prior art.

[0042] The application has good scalability and platform universality. The method of the application can be applied to various platforms such as wheeled robots, unmanned aerial vehicles, autonomous vehicles and collaborative robot arms, and can be migrated and deployed by only replacing the system dynamics model. Meanwhile, the system structure supports multi-robot cooperation and risk sharing, and realizes group safety control and cooperative obstacle avoidance.

[0043] To sum up, the application realizes risk perception, adaptive optimization and formal safety guarantee in an uncertain dynamic environment through a three-layer fusion mechanism of “learning-optimization-constraint”. The method uses a learning module to model the environment dynamics and risk distribution, and combines model predictive control (MPC) and control barrier function (CBF) to dynamically adjust the safety margin during the optimization process according to the environmental risk level, so as to balance the control performance and safety. Compared with traditional methods, the application has the following significant technical advantages: 1) risk perception and adaptability are enhanced, and the controller can automatically adjust the control strategy according to the environmental risk; 2) the introduction of the control barrier function constraint ensures that the system state is always in a safe and feasible range in theory, and has formal safety guarantee; 3) the control efficiency, optimized trajectory smoothness and energy consumption are significantly improved under the premise of ensuring safety, avoiding the problem of over-conservatism of traditional robust control; and 4) the robustness of the system is improved, so that it can still maintain stable navigation performance under the influence of perception noise, prediction error and environmental disturbance.

[0044] In one embodiment, as shown in Figure 3 Based on the above risk adaptive safety navigation method, the application also correspondingly provides a risk adaptive safety navigation system, which comprises: A vector generation module 100 is configured to obtain an environmental state vector at the current time based on the current state information of the robot, the motion information of the obstacles and the environmental information at a preset frequency; A risk assessment module 200 is configured to input the environmental state vector at the current time into a learning module for processing to obtain a risk assessment value; An adjustment module 300 is configured to adjust the safety constraint region of the control barrier function and the weight distribution of the risk weighted cost function in the model predictive control module based on the risk assessment value; An instruction generation module 400 is configured to use the model predictive control module to perform optimization and solving based on the adjusted risk weighted cost function and the safety constraint formed by the adjusted safety constraint region to obtain a target control instruction that satisfies the safety constraint and is optimal in cost; A driving module 500 is configured to send the target control instruction to the actuator of the robot to drive the robot to move.

[0045] In one embodiment, the system further comprises: a data collection unit configured to collect trajectory deviation, risk change data and safety constraint state after the robot executes as feedback data; a feedback unit configured to feed the feedback data back to the learning module and the model predictive control module to update the policy parameters of the learning module and the control strategy of the model predictive control module.

[0046] In an embodiment, the system further comprises: a first adjustment unit configured to decrease the safety constraint region if the risk evaluation value is higher than a preset first threshold value; a second adjustment unit configured to increase the safety constraint region if the risk evaluation value is lower than a preset second threshold value.

[0047] In an embodiment, the system further comprises: a third adjustment unit configured to increase the weight of the safety performance index in the risk weighted cost function in the model predictive control module if the risk evaluation value is higher than a preset first threshold value; a fourth adjustment unit configured to increase the weight of the motion efficiency index in the risk weighted cost function in the model predictive control module if the risk evaluation value is lower than a preset second threshold value.

[0048] In an embodiment, the system further comprises: a prediction unit configured to generate a prediction state corresponding to a multi-step future trajectory within a preset prediction time domain by using the model predictive control module; a verification unit configured to verify whether the prediction state satisfies the safety constraint constituted by the adjusted safety constraint region at each time within the prediction time domain; an instruction generation unit configured to retain the prediction trajectory branch satisfying the safety constraint at all times, and perform optimization calculation by combining the adjusted risk weighted cost function to obtain a target space instruction satisfying the safety constraint.

[0049] In an embodiment, the system further comprises: an iteration unit configured to iteratively perform the steps of generating an environment state vector, obtaining a risk evaluation value, generating a target control instruction, performing control, collecting feedback data and updating until a navigation task is completed.

[0050] Based on the above embodiments, the application further provides a terminal, and a structure diagram thereof can be as shown in Figure 4As shown. The terminal includes a processor, memory, network interface, and display screen connected via a device bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores operating devices and a risk-adaptive safety navigation program. The internal memory provides an environment for the operation of the operating devices and the risk-adaptive safety navigation program stored in the non-volatile storage medium. The network interface is used for communication with external terminals via a network connection. When executed by the processor, the risk-adaptive safety navigation program implements the steps of any of the aforementioned risk-adaptive safety navigation methods. The display screen can be a liquid crystal display (LCD) or an electronic ink display.

[0051] Those skilled in the art will understand that Figure 4 The structural schematic diagram shown is only a schematic diagram of a part of the structure related to the present invention solution, and does not constitute a limitation on the terminal on which the present invention solution is applied. The specific terminal may include more or fewer components than shown in the figure, or combine some components, or have different component arrangements.

[0052] In one embodiment, a terminal is provided, the terminal including a memory, a processor, and a risk-adaptive safety navigation program stored in the memory and executable on the processor. When the risk-adaptive safety navigation program is executed by the processor, it implements the steps of any risk-adaptive safety navigation method provided in the embodiments of the present invention.

[0053] This invention also provides a computer-readable storage medium storing a risk-adaptive safety navigation program. When executed by a processor, the risk-adaptive safety navigation program implements the steps of any risk-adaptive safety navigation method provided in this invention.

[0054] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0055] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above-mentioned division of each functional unit and module is exemplified, and in actual application, the above-mentioned functions can be completed by different functional units and modules according to needs, that is, the internal structure of the above-mentioned device is divided into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiment can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be realized in the form of hardware or software function unit. In addition, the specific name of each functional unit and module is only for the convenience of mutual distinction, and does not limit the protection scope of the present application. The specific working process of the units and modules in the above-mentioned device can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0056] In the above embodiments, the description of each embodiment has its own emphasis, and the parts not described or recorded in detail in a certain embodiment can be referred to the related description of other embodiments.

[0057] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0058] In the embodiments provided by the present application, it should be understood that the disclosed device / terminal equipment and method can be implemented by other ways. For example, the above-mentioned device / terminal equipment embodiments are only schematic, for example, the division of the above-mentioned modules or units is only a logical function division, and actual implementation can be in another division manner, for example, multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed.

[0059] The above-mentioned embodiments are only used to illustrate the technical solutions of the present application, but not limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand; the technical solutions recorded in the foregoing embodiments can be modified, or some technical features can be replaced by equivalents; and these modifications or replacements do not deviate from the spirit and scope of the corresponding technical solutions, and should be included in the protection scope of the present application.

Claims

1. A risk-adaptive safety navigation method, characterized in that, The method includes: The robot's current state information, obstacle motion information, and environmental information are acquired at a preset frequency to generate the current environmental state vector. The environmental state vector at the current moment is input into the learning module for processing to obtain the risk assessment value; Based on the risk assessment value, adjust the safety constraint region of the control barrier function and the weight allocation of the risk-weighted cost function in the model prediction control module; Using the model predictive control module, the target control command that satisfies both the safety constraints and the cost is optimized by solving the adjusted risk-weighted cost function and the safety constraints formed by the adjusted safety constraint region. The target control command is sent to the robot's actuator to drive the robot's movement.

2. The risk-adaptive safety navigation method according to claim 1, characterized in that, After sending the target control command to the robot's actuator to drive the robot's movement, the method further includes: Collect data on robot trajectory deviation, risk changes, and safety constraint status after execution as feedback data; The feedback data is sent back to the learning module and the model prediction control module to update the strategy parameters of the learning module and the control strategy of the model prediction control module.

3. The risk-adaptive safety navigation method according to claim 1, characterized in that, Based on the aforementioned risk assessment value, the safety constraint region of the control barrier function is adjusted, including: If the risk assessment value is higher than a preset first threshold, then the safety constraint area is reduced; If the risk assessment value is lower than the preset second threshold, the safety constraint area is increased.

4. The risk-adaptive safety navigation method according to claim 1, characterized in that, Based on the risk assessment value, the weight allocation of the risk-weighted cost function in the model prediction and control module is adjusted, including: If the risk assessment value is higher than the preset first threshold, then the weight of the safety performance index in the risk weighted cost function of the model prediction control module is increased. If the risk assessment value is lower than the preset second threshold, the weight of the motion efficiency index in the risk-weighted cost function of the model prediction control module is increased.

5. The risk-adaptive safety navigation method according to claim 1, characterized in that, The learning module includes a policy network trained based on a proximal policy optimization algorithm.

6. The risk-adaptive safety navigation method according to claim 1, characterized in that, The optimization process using the model predictive control module includes: The model prediction control module generates predicted states corresponding to multiple future trajectories within a preset prediction time domain. At each time point in the prediction time domain, the predicted state is verified to ensure that it meets the security constraints constituted by the adjusted security constraint region. The predicted trajectory branches that satisfy the safety constraints at all times are retained, and optimization calculations are performed using the adjusted risk-weighted cost function to obtain the target space instruction that satisfies the safety constraints and has the optimal cost.

7. The risk-adaptive safety navigation method according to claim 2, characterized in that, The method further includes: The process iteratively executes steps such as generating an environment state vector, obtaining risk assessment values, generating target control commands, executing control, collecting feedback data, and updating the data until the navigation task is completed.

8. A risk-adaptive safety navigation system, characterized in that, include: The vector generation module is used to acquire and generate the current environmental state vector based on the robot's current state information, obstacle motion information, and environmental information at a preset frequency. The risk assessment module inputs the current environmental state vector into the learning module for processing to obtain the risk assessment value; An adjustment module is used to adjust the safety constraint region of the control barrier function and the weight allocation of the risk-weighted cost function in the model prediction control module based on the risk assessment value. The instruction generation module is used to optimize the solution based on the adjusted risk-weighted cost function and the safety constraints formed by the adjusted safety constraint region using the model predictive control module, so as to obtain the target control instruction that satisfies the safety constraints and the optimal cost. The drive module is used to send the target control commands to the robot's actuators to drive the robot's movement.

9. A terminal, characterized in that, The terminal includes: a memory, a processor, and a risk-adaptive safety navigation program stored in the memory and executable on the processor, wherein when the risk-adaptive safety navigation program is executed by the processor, it implements the steps of the risk-adaptive safety navigation method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a risk-adaptive safety navigation program, which, when executed by a processor, implements the steps of the risk-adaptive safety navigation method as described in any one of claims 1-7.

Images