Robot autonomous obstacle avoidance control method and robot

Abstract

This invention provides a robot autonomous obstacle avoidance control method and robot, relating to the field of data processing technology. The method includes: acquiring environmental data of the robot and processing the environmental data to generate an environmental state description; calculating an environmental complexity value based on the environmental state description; and adjusting the exploration rate, learning rate, and reward function weights of the decision model in real time according to the environmental complexity value; wherein the exploration rate is determined based on the environmental complexity value and real-time performance feedback data; the learning rate is determined based on the environmental complexity value and the environmental change rate; and the safety weight and efficiency weight in the reward function weights are dynamically scaled based on the environmental complexity value and obstacle type; inputting the environmental state description into the adjusted decision model, outputting obstacle avoidance action commands, and generating control signals according to the obstacle avoidance action commands to control the robot's movement.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a robot autonomous obstacle avoidance control method and a robot. Background Technology

[0002] With the widespread application of artificial intelligence technology, robots, as intelligent mobile devices, are gradually integrating into daily life. Autonomous obstacle avoidance capability is a fundamental function to ensure the safe movement of robots, directly affecting their practicality and reliability in real-world environments.

[0003] When robots move autonomously in unstructured environments such as homes, they need to generate obstacle avoidance commands based on environmental information acquired by sensors to avoid collisions with static or dynamic obstacles. Existing obstacle avoidance control schemes often employ rule-based methods that rely on fixed thresholds and preset logic, which can easily lead to policy mismatch when faced with changes in environmental layout or dynamic obstacles. Machine learning-based methods mostly use decision models with fixed parameters trained offline, making it difficult to adjust decision-making behavior synchronously with fluctuations in environmental complexity and change after deployment. This can result in unstable obstacle avoidance safety and movement efficiency performance under different complexity scenarios.

[0004] Therefore, how to more effectively control robot autonomous obstacle avoidance has become an urgent problem to be solved in the industry. Summary of the Invention

[0005] This invention provides a robot autonomous obstacle avoidance control method and robot, to solve the problem of how to more effectively control robot autonomous obstacle avoidance in the prior art.

[0006] This invention provides a robot autonomous obstacle avoidance control method, comprising: Acquire environmental data of the robot and process the environmental data to generate an environmental state description; The environmental complexity value is calculated based on the environmental state description. The exploration rate, learning rate, and reward function weights of the decision model are adjusted in real time according to the environmental complexity value. The exploration rate is determined based on the environmental complexity value and real-time performance feedback data. The learning rate is determined based on the environmental complexity value and the environmental change rate. The safety weight and efficiency weight in the reward function are dynamically scaled based on the environmental complexity value and obstacle type. The environmental state description is input into the adjusted decision model, which outputs obstacle avoidance action commands. Control signals are generated based on the obstacle avoidance action commands to control the robot's movement.

[0007] According to a robot autonomous obstacle avoidance control method provided by the present invention, the method for determining the exploration rate includes: The environmental complexity value and real-time performance feedback data are obtained; wherein, the real-time performance feedback data includes the number of consecutive successful obstacle avoidances and near-collision events of the robot; Based on the environmental complexity value, a basic exploration rate is calculated using a nonlinear mapping function; wherein, the basic exploration rate increases with the increase of the environmental complexity value; The exploration rate adjustment is calculated based on the real-time performance feedback data, and the exploration rate is obtained by summing the base exploration rate and the exploration rate adjustment.

[0008] The nonlinear mapping function is the Sigmoid function; the basic exploration rate is in the first exploration rate interval when the environmental complexity value is less than the first complexity threshold, and in the second exploration rate interval when the environmental complexity value is greater than the second complexity threshold, and the value of the second exploration rate interval is greater than the value of the first exploration rate interval; When the number of consecutive successful obstacle avoidances reaches a preset threshold, the exploration rate adjustment amount is increased; when a near-collision event is detected, the exploration rate adjustment amount is decreased.

[0009] According to a robot autonomous obstacle avoidance control method provided by the present invention, the method for determining the learning rate includes: Based on the environmental complexity value, the decay factor is calculated using the exponential decay function; Based on the environmental change rate, the stability factor is calculated using the stability compensation function. The learning rate is obtained based on the base learning rate, the decay factor, and the stability factor.

[0010] The method for determining the learning rate specifically includes: The environmental change rate is obtained; wherein the environmental change rate is calculated based on the difference norm of sensor readings between consecutive frames; The step of calculating the decay factor based on the environmental complexity value using an exponential decay function includes: calculating the decay factor so that the decay factor decreases as the environmental complexity value increases; The step of calculating the stability factor based on the environmental change rate using a stability compensation function includes: calculating the stability factor so that the stability factor decreases as the environmental change rate increases.

[0011] According to a robot autonomous obstacle avoidance control method provided by the present invention, the method for determining the safety weight and efficiency weight in the reward function weight includes: The complexity scaling factor is determined based on the aforementioned environment complexity value; Identify the types of obstacles in the environmental state description and determine the corresponding type coefficients; wherein, the type coefficients for dynamic obstacles are greater than those for static obstacles; The security weights are obtained by positively scaling the security base weights based on the complexity scaling factor and the type coefficient. Based on the complexity scaling factor, the efficiency base weight is inversely scaled to obtain the efficiency weight. The reward function of the decision model is the product of the safety weight and the safety reward, the product of the efficiency weight and the efficiency reward, and the weighted sum of the target reward.

[0012] According to the present invention, a robot autonomous obstacle avoidance control method calculates an environmental complexity value based on the environmental state description, including: Based on the environmental state description, the obstacle density, number of dynamic objects, environmental change rate, and proportion of passable area are calculated. The environmental complexity value is obtained by weighted summation of the obstacle density, the number of dynamic objects, the rate of environmental change, and the proportion of passable areas, and then normalizing the summation result to a preset interval.

[0013] According to the present invention, a robot autonomous obstacle avoidance control method is provided, wherein acquiring the robot's environmental data includes: The robot collects data about its surrounding environment using a set of sensors installed on it. Real-time monitoring of the rate of environmental change; wherein the rate of environmental change is obtained by calculating the difference norm of sensor readings between consecutive frames; The sampling frequency of the sensor array is dynamically adjusted according to the environmental change rate; wherein the sampling frequency is positively correlated with the environmental change rate.

[0014] According to the present invention, a robot autonomous obstacle avoidance control method generates control signals based on the obstacle avoidance action command, including: By using the high-level trajectory planning layer in the hierarchical control architecture, cubic spline interpolation is used to convert the path points corresponding to the obstacle avoidance action command into a smooth path. The bottom tracking control layer in the hierarchical control architecture uses a controller based on the proportional-integral-derivative algorithm to track the smooth path and output the control signal. When generating the control signal, a feedforward compensation term is introduced to offset the system inertial delay.

[0015] The present invention also provides a robot, comprising: A sensor array is used to collect environmental data around the robot; Memory, used to store computer programs; A processor is configured to execute the computer program to perform the following steps: The system acquires environmental data collected by the sensor array, processes the environmental data, and generates an environmental state description. The environmental complexity value is calculated based on the environmental state description. The exploration rate, learning rate, and reward function weights of the decision model are adjusted in real time according to the environmental complexity value. The exploration rate is determined based on the environmental complexity value and real-time performance feedback data. The learning rate is determined based on the environmental complexity value and the environmental change rate. The safety weight and efficiency weight in the reward function are dynamically scaled based on the environmental complexity value and obstacle type. The environmental state description is input into the adjusted decision model, which outputs obstacle avoidance action instructions and generates control signals based on the obstacle avoidance action instructions. An actuator is used to respond to the control signal to drive the robot to move.

[0016] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the robot autonomous obstacle avoidance control method described above.

[0017] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the robot autonomous obstacle avoidance control method as described above.

[0018] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the robot autonomous obstacle avoidance control method as described above.

[0019] This invention provides a robot autonomous obstacle avoidance control method and robot. It acquires environmental data and generates an environmental state description, calculates the environmental complexity value based on the environmental state description, and adjusts the exploration rate, learning rate, and reward function weights of the decision model in real time according to this environmental complexity value. The exploration rate is determined by a nonlinear mapping function combined with real-time performance feedback data; the learning rate is determined by an exponential decay function combined with a stability compensation function and the rate of environmental change; and the safety and efficiency weights in the reward function are dynamically scaled based on the environmental complexity value and obstacle type. The environmental state description is input into the adjusted decision model, which outputs obstacle avoidance action commands to control the robot's movement. This establishes a dynamic mapping relationship between environmental complexity and decision parameters. In low-complexity environments, the decision model reduces the exploration rate to utilize existing strategies and improve movement efficiency. In high-complexity environments, it increases the exploration rate and safety weights to enhance the exploration of new strategies and prioritize safety. Simultaneously, the safety weights are dynamically adjusted according to obstacle type to provide a higher safety margin for dynamic obstacles. This achieves adaptive obstacle avoidance control without human intervention, effectively solving the problem that fixed-parameter obstacle avoidance control systems cannot adapt to dynamic environmental changes. It improves the robot's movement efficiency in different complexity environments while ensuring obstacle avoidance safety. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0021] Figure 1 This is a flowchart illustrating the robot autonomous obstacle avoidance control method provided by the present invention; Figure 2 This is a schematic diagram of the robot structure provided by the present invention; Figure 3 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0023] The robot autonomous obstacle avoidance control method provided in this invention can be applied to robots with autonomous mobility, including but not limited to pet robots, service robots, and cleaning robots. The execution subject of this method can be a controller, processor, or control system built into the robot, or an external computing device communicatively connected to the robot. In the following embodiments, the controller built into a pet robot is used as the execution subject for illustration.

[0024] Figure 1 This is a flowchart illustrating the robot autonomous obstacle avoidance control method provided by the present invention, as shown below. Figure 1 As shown, the method includes the following: Step 110: Obtain the robot's environmental data and process the environmental data to generate an environmental state description; In this embodiment, the environmental data refers to the raw data information about the surrounding environment collected by the robot through its onboard sensors. This environmental data may include, but is not limited to: distance information between obstacles and the robot, location distribution information of obstacles, motion state information of objects in the environment, and visual feature information of the environment.

[0025] Environmental data for a robot can be acquired through a sensor array installed on the robot. This sensor array can include one or more combinations of various types of sensors, such as lidar, cameras, ultrasonic sensors, and infrared sensors.

[0026] Different types of sensors have their own sensing characteristics. LiDAR can provide precise distance measurement data; cameras can acquire rich visual feature information; and ultrasonic sensors are suitable for detecting transparent objects or obstacles made of special materials. By working together with multiple sensors, more comprehensive and accurate environmental data can be obtained.

[0027] In a specific application scenario, the operating environment of a pet robot is usually a home environment, which contains various types of obstacles.

[0028] These obstacles can be categorized into two types based on their motion characteristics: static obstacles, which include stationary objects such as walls, furniture, and door frames; and dynamic obstacles, which include moving objects such as walking humans and moving pets. The sensor array needs to be able to detect and distinguish between these different types of obstacles.

[0029] Environmental data processing refers to preprocessing the collected raw environmental data to improve data quality and standardize data format. This process may include steps such as data cleaning, noise filtering, and data standardization.

[0030] Data cleaning is used to remove outliers or erroneous data generated during sensor acquisition; noise filtering uses filtering algorithms to smooth the data and reduce the impact of noise interference; data standardization is used to convert heterogeneous data collected by different sensors into a unified format to ensure consistency in subsequent processing.

[0031] An environmental state description is structured information obtained by comprehensively representing processed environmental data. This environmental state description can reflect the overall characteristics of the current environment, including the spatial distribution of obstacles, the range of passable areas, and the motion state of dynamic objects.

[0032] Environmental state descriptions can be represented using vectors, matrices, or other data structures to facilitate subsequent calculations and processing.

[0033] Step 120: Calculate the environmental complexity value based on the environmental state description, and adjust the exploration rate, learning rate, and reward function weights of the decision model in real time according to the environmental complexity value; wherein, the exploration rate is determined based on the environmental complexity value and real-time performance feedback data; the learning rate is determined based on the environmental complexity value and the environmental change rate; the safety weight and efficiency weight in the reward function weights are dynamically scaled based on the environmental complexity value and obstacle type. In this embodiment, the environment complexity value is a numerical indicator used to quantify the complexity of the current environment.

[0034] The environmental complexity value can comprehensively reflect the degree of challenge that the robot's environment poses to obstacle avoidance decisions. Its value is usually normalized to a preset range (e.g., the range [0,1]) to facilitate subsequent parameter adjustment calculations.

[0035] In calculating the environmental complexity value based on the environmental state description, several factors that affect the degree of environmental complexity can be considered. These factors may include the density of obstacles, the number and motion characteristics of dynamic objects, the rate of change of the environment, and the proportion of areas accessible to the robot.

[0036] By comprehensively evaluating these factors, an environmental complexity value can be obtained that fully reflects the degree of environmental complexity.

[0037] The decision model is used to generate obstacle avoidance decisions based on environmental states. This model can be implemented using a reinforcement learning framework and includes components such as a state encoding layer, a policy network, and a value evaluation network. The state encoding layer converts the environmental state description into feature vectors; the policy network outputs the action probability distribution; and the value evaluation network predicts long-term rewards.

[0038] The decision model can be initialized through offline training and updated online based on real-time data during actual operation.

[0039] The exploration rate is a parameter used in decision-making models to control the balance between exploration and exploitation. In reinforcement learning frameworks, the exploration rate determines the probability that a robot will attempt new actions when making decisions. A higher exploration rate makes the robot more inclined to try unknown action strategies, which is beneficial for discovering better obstacle avoidance solutions; a lower exploration rate makes the robot more inclined to utilize learned experience, which is beneficial for maintaining the stability of decisions.

[0040] In this embodiment, the exploration rate is determined by a nonlinear mapping function based on the environmental complexity value and real-time performance feedback data.

[0041] Nonlinear mapping functions can establish a nonlinear correspondence between environmental complexity values ​​and exploration rates, enabling the exploration rate to be adaptively adjusted according to the degree of environmental complexity.

[0042] Specifically, when the environmental complexity value is high (e.g., the robot is in a narrow space with dense obstacles), the exploration rate increases accordingly, prompting the robot to try more obstacle avoidance strategies to cope with the complex environment; when the environmental complexity value is low (e.g., the robot is in an open area), the exploration rate decreases accordingly, allowing the robot to make more use of existing successful experiences for efficient movement.

[0043] Real-time performance feedback data refers to the evaluation data of the robot's actual performance during operation, which may include indicators such as obstacle avoidance success rate and collision event occurrence.

[0044] By combining real-time performance feedback data, the adjustment of the exploration rate can better adapt to the actual operating state of the robot, avoiding the adjustment deviation that may be caused by relying on a single indicator of environmental complexity.

[0045] The learning rate is a parameter used in decision models to control the magnitude of parameter updates. It determines the extent to which the model absorbs newly acquired information with each update. A higher learning rate allows the model to adapt quickly to environmental changes but may lead to training instability; a lower learning rate makes model updates more stable but results in a slower response to environmental changes.

[0046] In this embodiment, the learning rate is determined based on the environmental complexity value and the environmental change rate, using an exponential decay function and a stability compensation function. The environmental change rate is an indicator used to describe the drastic degree of dynamic change in the environment.

[0047] The exponential decay function is used to calculate the decay factor based on the environmental complexity value, so that the learning rate decreases appropriately as the environmental complexity increases, thereby improving the learning stability in complex environments.

[0048] The stability compensation function is used to further adjust the learning rate according to the rate of change of the environment. When the environment changes drastically, the learning rate is reduced to avoid over-updating, and when the environment is relatively stable, an appropriate learning rate is maintained to ensure learning efficiency.

[0049] Reward function weights are parameters used in decision-making models to balance different optimization objectives. In obstacle avoidance tasks, robots need to consider multiple objectives simultaneously, such as safety and efficiency. The reward function weights determine the relative importance of different objectives in the overall reward calculation.

[0050] In this embodiment, the safety weight and efficiency weight in the reward function weight are dynamically scaled and determined based on the environmental complexity value and obstacle type.

[0051] Safety weight corresponds to the weight of obstacle avoidance safety objectives, while efficiency weight corresponds to the weight of mobility efficiency objectives. When the environmental complexity is high or there are dynamic obstacles, the safety weight increases accordingly while the efficiency weight decreases accordingly, making the robot focus more on obstacle avoidance safety. When the environmental complexity is low and the obstacles are mainly static, the safety weight decreases appropriately while the efficiency weight increases accordingly, enabling the robot to move more efficiently.

[0052] Step 130: Input the environmental state description into the adjusted decision model, output obstacle avoidance action instructions, and generate control signals according to the obstacle avoidance action instructions to control the movement of the robot.

[0053] In this embodiment, after adjusting the parameters of the decision model, the controller inputs the current environmental state description into the adjusted decision model. Based on the input environmental state description and the adjusted parameter configuration, the decision model performs inference calculations through its internal policy network and outputs corresponding obstacle avoidance action commands.

[0054] Obstacle avoidance commands are output by the decision model to guide the robot in obstacle avoidance movements. These commands can take the form of discrete actions, such as basic actions like forward, backward, left turn, right turn, and stop; or they can take the form of continuous actions, such as specifying specific values ​​for linear velocity and angular velocity.

[0055] The form of obstacle avoidance commands can be selected based on the specific type of robot and the characteristics of the control system.

[0056] Generating control signals from obstacle avoidance commands refers to converting high-level obstacle avoidance commands into low-level control signals that can directly drive the robot's actuators. This conversion process needs to consider the robot's kinematic characteristics and dynamic constraints to ensure that the generated control signals are physically feasible and enable the robot to safely and smoothly execute the intended obstacle avoidance actions.

[0057] Control signals are used to drive the robot's actuators, such as drive motors and servo motors, thereby controlling the robot to move according to the requirements of obstacle avoidance instructions.

[0058] By executing control signals, the robot can perform movement behaviors such as turning, accelerating, decelerating, and stopping, ultimately completing the obstacle avoidance task.

[0059] More specifically, during the obstacle avoidance action command generation process, the decision model comprehensively considers both action accuracy and physical constraints to ensure that the output obstacle avoidance action commands not only meet obstacle avoidance requirements but also have physical feasibility. Home environments typically involve various types of movement tasks, and robots need to perform diverse movement behaviors such as moving forward, turning, and stopping in different scenarios. Therefore, the design of obstacle avoidance action commands needs to balance flexibility and accuracy.

[0060] Obstacle avoidance commands can be defined in either discrete action space or continuous action space.

[0061] The discrete action space defines several basic action categories, including forward, backward, left turn, right turn, stationary rotation, and stop. These basic actions cover common movement patterns of robots in the home environment. The decision model selects the action to be performed by outputting the probability distribution of the action category.

[0062] Continuous motion space, through a regression network, directly outputs precise control command values, such as the specific values ​​of linear and angular velocities, enabling the robot to perform more refined and smooth motion adjustments and achieve silky obstacle avoidance trajectories. In practical applications, the appropriate motion space form can be selected according to task requirements and control accuracy requirements, or a combination of the two forms can be used.

[0063] The generation of actions strictly adheres to kinematic constraints to ensure that the generated obstacle avoidance commands are physically feasible. This invention uses a bicycle model to model the robot's motion characteristics, which can accurately describe the motion laws of wheeled robots.

[0064] The kinematic equations of the bicycle model are: , , ,in, ẋ and ẏ The robots are respectively in x direction and y The velocity component in the direction, θ The robot's heading angle, The rate of change of the heading angle, v Let the linear velocity of the robot be . For the robot's wheelbase, This is the steering angle.

[0065] This model uses linear velocity v and steering angle δ Two control inputs describe the changes in the robot's motion state, accurately reflecting the robot's trajectory on the plane and ensuring that the trajectory generated by the decision model is physically feasible.

[0066] To ensure the safety and stability of the robot's movement, the generation of obstacle avoidance commands must also meet dynamic constraints. The maximum acceleration limit is set at 1.2 m / s². 2 This limit takes into account the characteristics of common floor materials in the home environment and the output capacity of the robot's power system. It can ensure that the robot has sufficient mobility to cope with sudden obstacle avoidance needs, while avoiding slippage or loss of control due to excessive acceleration.

[0067] The maximum angular velocity is set to 1.0 rad / s. This limit prevents the robot from tipping over or becoming unstable due to excessive centrifugal force during rapid turns, ensuring good attitude stability when performing turning maneuvers. By introducing these kinematic and dynamic constraints, the obstacle avoidance commands output by the decision model can ensure both obstacle avoidance effectiveness and the safety and smoothness of the robot's motion.

[0068] In this embodiment of the invention, the exploration rate, learning rate, and reward function weights of the decision model can be adjusted in real time according to the environmental complexity value, enabling the robot's obstacle avoidance strategy to adaptively match the current environmental characteristics. When the environment is complex, the exploration rate is increased to promote the discovery of new strategies, the learning rate is reduced to ensure learning stability, and safety weights are increased to prioritize obstacle avoidance safety. When the environment is simple, the system utilizes existing experience by reducing the exploration rate, improves learning efficiency by maintaining an appropriate learning rate, and enhances movement efficiency by increasing efficiency weights. This dynamic adjustment mechanism overcomes the shortcomings of fixed-parameter models in existing technologies that cannot adapt to environmental changes, achieving adaptive obstacle avoidance control without human intervention, and significantly improving the robot's movement efficiency while ensuring safety.

[0069] Optionally, the method for determining the exploration rate includes: The environmental complexity value and real-time performance feedback data are obtained; wherein, the real-time performance feedback data includes the number of consecutive successful obstacle avoidances and near-collision events of the robot; Based on the environmental complexity value, a basic exploration rate is calculated using a nonlinear mapping function; wherein, the basic exploration rate increases with the increase of the environmental complexity value; The exploration rate adjustment is calculated based on the real-time performance feedback data, and the exploration rate is obtained by summing the base exploration rate and the exploration rate adjustment.

[0070] In this application, the environmental complexity value is obtained as described in the foregoing embodiments. It is a quantitative index calculated based on the environmental state description, which can reflect the complexity of the current environment.

[0071] The real-time performance feedback data is an evaluation of the robot's performance during actual operation. In this embodiment, the real-time performance feedback data includes two types of information: the number of consecutive successful obstacle avoidances by the robot and near-collision events.

[0072] The consecutive successful obstacle avoidance count refers to the number of times the robot successfully completes obstacle avoidance maneuvers consecutively without a collision. This metric reflects the effectiveness of the current obstacle avoidance strategy: a high consecutive successful obstacle avoidance count indicates that the current strategy performs well in this environment; a low consecutive successful obstacle avoidance count or frequent resets indicate that the current strategy may have shortcomings.

[0073] A near-collision event refers to a situation where the distance between the robot and an obstacle is less than a preset safety threshold, but an actual collision has not yet occurred. The occurrence of a near-collision event indicates that the robot faces certain risks during obstacle avoidance, and the safety margin of the current strategy may be insufficient. By monitoring near-collision events, the system can adjust the exploration rate in a timely manner to improve obstacle avoidance safety.

[0074] A nonlinear mapping function is used to map environmental complexity values ​​to a base exploration rate; that is, the base exploration rate is the output of the nonlinear mapping function. The reason for using a nonlinear mapping function instead of a nonlinear function is that the relationship between environmental complexity and exploration rate is not a simple linear proportional relationship.

[0075] In the low and high complexity ranges, the exploration rate should change gradually; while in the medium complexity range, the exploration rate should show high sensitivity to changes in complexity.

[0076] Nonlinear mapping functions can satisfy this adjustment requirement of saturation at both ends and sensitivity in the middle, ensuring that the base exploration rate is at its minimum. With the maximum value The values ​​are adjusted smoothly and non-linearly according to the environmental complexity.

[0077] The final exploration rate is obtained by adjusting the base exploration rate. With the exploration rate adjustment Δ The results are added together and then truncated to ensure the exploration rate remains within a valid range. Inside.

[0078] The embodiments of the present invention use a combination mechanism of basic exploration rate and exploration rate adjustment to enable the exploration rate to reflect the characteristics of environmental complexity and adapt to the actual operating performance of the robot, thereby maintaining a good balance between exploration and utilization in different scenarios.

[0079] Optionally, the nonlinear mapping function is a Sigmoid function; the basic exploration rate is in a first exploration rate interval when the environmental complexity value is less than a first complexity threshold, and in a second exploration rate interval when the environmental complexity value is greater than a second complexity threshold, wherein the value of the second exploration rate interval is greater than the value of the first exploration rate interval; When the number of consecutive successful obstacle avoidances reaches a preset threshold, the exploration rate adjustment amount is increased; when a near-collision event is detected, the exploration rate adjustment amount is decreased.

[0080] In this application, the Sigmoid function is a specific implementation of the nonlinear mapping function. It has S-shaped curve characteristics, and its output value is between 0 and 1, which can realize a smooth nonlinear mapping from the environment complexity value to the basic exploration rate.

[0081] When using the Sigmoid function, the formula for calculating the base exploration rate is: ; in, Based on the exploration rate, and These are the minimum and maximum values ​​of the exploration rate, respectively. The slope parameter is used to control the sensitivity of the adjustment. This represents the environmental complexity value. This is the offset for environmental complexity. Adjust the parameters... and It can control the response characteristics of the base exploration rate as environmental complexity changes.

[0082] When the environmental complexity value C Much smaller than the offset C 0 At this point, corresponding to the first complexity threshold, the output value of the Sigmoid function approaches 0, and the basic exploration rate approaches 0. It is in the first exploration rate range; when the environmental complexity value C Much larger than the offset C 0 At the second complexity threshold, the output value of the Sigmoid function approaches 1, and the basic exploration rate approaches 1. It is in the second exploration rate range.

[0083] The first complexity threshold is used to define low-complexity environments. When the environment complexity value is less than the first complexity threshold, the current environment is considered relatively simple, such as a robot in an open living room area with sparse and mostly static obstacles. In this case, the base exploration rate is in the first exploration rate range, which has lower values, allowing the robot to make more use of learned effective strategies for efficient movement.

[0084] The second complexity threshold is used to define high-complexity environments. When the environmental complexity value exceeds the second complexity threshold, the current environment is considered relatively complex. For example, the robot may be in a narrow corridor or an area with dense furniture, numerous obstacles, or many dynamic objects. In this case, the base exploration rate falls within the second exploration rate range, which has a higher value, prompting the robot to try more obstacle avoidance strategies to cope with the complex environment.

[0085] The exploration rate adjustment is a correction to the base exploration rate calculated based on the robot's real-time performance. By introducing the exploration rate adjustment, the determination of the exploration rate not only considers the static factor of environmental complexity but also responds to the robot's dynamic operating state, enabling more refined adjustments.

[0086] In this embodiment, when the number of consecutive successful obstacle avoidances reaches a preset threshold, the exploration rate adjustment is increased. This is because consecutive successful obstacle avoidances indicate that the current strategy is relatively mature, and appropriately increasing the exploration rate helps prevent the strategy from converging to a local optimum too early, promoting the discovery of potentially better strategies.

[0087] When a near-collision event is detected, the exploration rate adjustment is reduced. This is because a near-collision event indicates that the current strategy has insufficient safety margins; reducing the exploration rate allows the robot to rely more on existing safety strategies, thus reducing the risk of collision.

[0088] As a specific implementation method, the exploration rate adjustment amount The calculation can be based on the following rule: when the number of consecutive successful obstacle avoidances reaches a preset threshold (e.g., 15 times), Increase the preset increment value (e.g., 0.05); when a near collision event is detected, Reduce the preset reduction value (e.g., 0.1). The exploration rate adjustment can be constrained by upper and lower limits to avoid excessive adjustment that could lead to system instability.

[0089] Final exploration rate By base exploration rate Adjustment amount with exploration rate The sum is obtained by addition; to ensure the effectiveness of the exploration rate, the summation result can be truncated to keep it within the effective range.

[0090] This invention enables fine-grained adaptive adjustment of the exploration rate. The base exploration rate is calculated using the Sigmoid function based on environmental complexity, allowing for a smooth transition from low to high complexity environments. The exploration rate adjustment is calculated based on real-time performance feedback data, responding to the robot's actual operating state. The combination of these two factors ensures that the final exploration rate reflects both environmental characteristics and adapts to the robot's performance, maintaining a good balance between exploration and utilization in different scenarios and improving the overall performance of obstacle avoidance control.

[0091] Optionally, the method for determining the learning rate includes: Based on the environmental complexity value, the decay factor is calculated using the exponential decay function; Based on the environmental change rate, the stability factor is calculated using the stability compensation function. The learning rate is obtained based on the base learning rate, the decay factor, and the stability factor.

[0092] In this application, the environmental complexity value is obtained as described in the foregoing embodiments. The environmental change rate is an indicator used to describe the degree of drastic dynamic changes in the environment.

[0093] In this embodiment, the exponential decay function is a function used to map the environmental complexity value to a decay factor, that is, the decay factor is the output of the exponential decay function.

[0094] The mathematical form of the exponential decay function is: ; in, As the attenuation factor, η This is the attenuation coefficient (a positive number). C Let be the environmental complexity value. This formula causes the decay factor to decrease exponentially with increasing environmental complexity value.

[0095] The decay factor decreases exponentially with increasing environmental complexity, ranging from (0, 1). C=0 hour, The learning rate is unaffected by decay; as C The increase, The learning rate rapidly approaches zero, resulting in a significant decrease. The reason for using exponential decay instead of nonlinear decay is that exponential decay provides stronger inhibition in highly complex environments, reducing the learning rate more quickly to ensure learning stability when the environment is complex.

[0096] The stability compensation function is a function used to map the rate of environmental change to a stability factor; that is, the stability factor is the output of the stability compensation function. The mathematical form of the stability compensation function is: ; in, As a stability factor, β The stability coefficient (is a positive number). V Let be the rate of environmental change. This formula makes the stability factor decrease monotonically as the rate of environmental change increases. The value range is (0, 1).

[0097] When the rate of environmental change V=0 =1, the learning rate is not affected by stability compensation; as V increases, As the learning rate gradually decreases, the learning rate also decreases accordingly. The purpose of introducing a stability compensation function is to further reduce the learning rate when the environment changes drastically, thereby preventing the model from over-responding to instantaneous changes and causing unnecessary parameter oscillations.

[0098] The exponential decay function and the stability compensation function adjust the learning rate from two dimensions: environmental complexity and environmental dynamics, respectively. Their synergistic effect enables the learning rate to better adapt to complex and ever-changing environmental conditions.

[0099] The final learning rate α is determined by the base learning rate. With decay factor Stability factor Multiplying them together gives: ; in, The base learning rate is a pre-set initial value for the learning rate.

[0100] The embodiments of the present invention employ a two-factor adjustment mechanism of exponential decay function and stability compensation function to enable the learning rate to automatically decrease in complex or rapidly changing environments to improve stability, and to maintain an appropriate level in simple or stable environments to improve efficiency.

[0101] Optionally, the method for determining the learning rate specifically includes: The environmental change rate is obtained; wherein the environmental change rate is calculated based on the difference norm of sensor readings between consecutive frames; The step of calculating the decay factor based on the environmental complexity value using an exponential decay function includes: calculating the decay factor so that the decay factor decreases as the environmental complexity value increases; The step of calculating the stability factor based on the environmental change rate using a stability compensation function includes: calculating the stability factor so that the stability factor decreases as the environmental change rate increases.

[0102] Environmental change rate V It is calculated based on the difference norm of sensor readings between consecutive frames, and the specific calculation formula is as follows: ; in Number of sensor data points For a moment Sensor readings, For a moment t-1 The i Each sensor reading. This formula calculates the average of the absolute differences between all sensor readings between two consecutive frames, reflecting the degree of environmental change over a short period of time.

[0103] This formula calculates the average of the absolute differences between all sensor readings between two consecutive frames, reflecting the degree of environmental change over a short period. When there are fast-moving objects in the environment or the robot itself is moving rapidly, the rate of environmental change... V The value is relatively high; when the environment is relatively stable or the robot moves slowly, the rate of environmental change is low. V The value is low.

[0104] Based on the above environmental change rate V According to the calculation results, the decay factor in the aforementioned exponential decay function decreases as the environmental complexity value increases.

[0105] As can be seen from the monotonically decreasing property of exponential functions, when the environmental complexity value C When the environmental complexity value is small, the decay factor is close to 1, and the learning rate remains basically at the base learning rate level; when the environmental complexity value is small... C As the attenuation factor increases, the learning rate decreases rapidly, thus improving learning stability in high-complexity environments. (Attenuation coefficient) η The decay rate was controlled. η The larger the value, the more sensitive the attenuation factor becomes to changes in environmental complexity.

[0106] The stability factor in the aforementioned stability compensation function decreases as the rate of environmental change increases. Due to the monotonically decreasing property of this function, when the rate of environmental change... V When the environmental change rate is small, the stability factor is close to 1, and the learning rate is not significantly affected; when the environmental change rate is small... V As the stability factor increases, the learning rate decreases gradually, further reducing on the basis of exponential decay. This prevents the model from over-responding to instantaneous changes when the environment changes drastically. (Stability coefficient) β The intensity of compensation was controlled. β The larger the value, the more sensitive the stability factor is to the rate of environmental change.

[0107] This invention provides a precise means of quantifying the dynamic characteristics of the environment by calculating the rate of change of the environment using the difference norm; by further defining the characteristics of the change of the decay factor and the stability factor, the specific response behavior of the learning rate under different environmental conditions is clarified.

[0108] Optionally, the complexity scaling factor is a coefficient used to adjust the weights, calculated based on the environment complexity value. The complexity scaling factor can pass the influence of environment complexity into the weight calculation process, allowing the weights to adjust as the environment complexity changes.

[0109] The complexity scaling factor can be calculated linearly or non-linearly. As a specific implementation, the complexity scaling factor can be a linear function of the environment complexity value: ; This is the complexity scaling factor. This is the scaling factor (a positive number). This represents the environmental complexity value. The formula makes the complexity scaling factor proportional to the environmental complexity value.

[0110] Obstacle types are classifications of obstacles in the environment based on specific attributes. In this embodiment, obstacle types include at least two categories: static obstacles and dynamic obstacles.

[0111] Static obstacles refer to obstacles whose positions remain unchanged, such as walls, furniture, and door frames. The movement trajectory of static obstacles is predictable, and the robot can plan a stable obstacle avoidance path based on their fixed positions.

[0112] Dynamic obstacles are those whose positions change over time, such as walking people or moving pets. The trajectories of dynamic obstacles are unpredictable, requiring robots to monitor their positional changes in real time and adjust their obstacle avoidance strategies accordingly, making obstacle avoidance relatively challenging.

[0113] The type coefficient is a weight adjustment coefficient corresponding to the obstacle type. In this embodiment, the type coefficient for dynamic obstacles is greater than that for static obstacles. This design reflects the higher obstacle avoidance risk posed by dynamic obstacles: due to the uncertainty of the position of dynamic obstacles, the risk of collision with them is higher, therefore a larger type coefficient is needed to increase the safety weight.

[0114] As one specific implementation method, This is the obstacle type coefficient, for static obstacles. Dynamic obstacles .

[0115] When multiple types of obstacles exist in the environment, the type coefficient can be the maximum value of the corresponding coefficients for each type of obstacle, or it can be calculated by weighting the coefficients according to the proportion of each type of obstacle.

[0116] The basic safety weight is the initial setting of the safety weight, which is the default safety weight without considering the influence of environmental complexity and obstacle type.

[0117] The forward scaling refers to an adjustment method where the safety weight increases accordingly as the scaling factor and type coefficient increase. The principle behind this design is that when the environment is complex or dynamic obstacles are present, the importance of obstacle avoidance safety increases, requiring an increase in safety weight to make the decision model pay more attention to obstacle avoidance safety.

[0118] The safety weight can be calculated using the following formula: ; in, For safety weights, For security-based weights, This is the complexity scaling factor. For obstacle type coefficients, this formula increases the safety weight as environmental complexity and obstacle risk level increase.

[0119] The basic efficiency weight is the initial setting of the efficiency weight, which is the default efficiency weight when the influence of environmental complexity is not considered.

[0120] Reverse scaling refers to an adjustment method where the efficiency weight decreases as the scaling factor increases. The principle behind this design is that when the environment is highly complex, ensuring obstacle avoidance safety becomes the primary goal, and the pursuit of movement efficiency should be appropriately reduced, thus requiring a decrease in the efficiency weight; when the environment is less complex, movement efficiency can be pursued more while ensuring safety, so the efficiency weight can remain higher.

[0121] The efficiency weight can be calculated using the following formula: ; in, For efficiency weighting, As the basic weight for efficiency, a lower limit constraint can be imposed on the calculation results to avoid negative values ​​for efficiency weights.

[0122] In this embodiment, the reward function of the decision model is the product of the safety weight and the safety reward, the product of the efficiency weight and the efficiency reward, and the weighted sum of the target reward.

[0123] Specifically, the reward function R can be expressed as: Among them, security weight and efficiency weight Based on environmental complexity And obstacle type dynamically scales, This is a safety reward item used to assess obstacle avoidance safety. A positive reward is given when the robot maintains a safe distance from obstacles, and a negative reward is given when a collision or near-collision occurs. This is an efficiency reward item used to evaluate movement efficiency; a positive reward is given when the robot moves quickly toward the target. This is a target reward item used to evaluate the achievement of the target; a positive reward is given when the robot reaches the target location. The target weight.

[0124] This invention enables the dynamic adjustment of the reward function's weights based on environmental complexity and obstacle type. In complex environments or scenarios with dynamic obstacles, safety weights increase while efficiency weights decrease, prioritizing obstacle avoidance safety for the robot. In simpler environments, safety weights decrease appropriately while efficiency weights increase, allowing the robot to move more efficiently. This dynamic weight adjustment mechanism allows the decision-making model to automatically balance safety and efficiency in different scenarios, achieving adaptive obstacle avoidance control.

[0125] Optionally, calculating the environmental complexity value based on the environmental state description includes: Based on the environmental state description, the obstacle density, number of dynamic objects, environmental change rate, and proportion of passable area are calculated. The environmental complexity value is obtained by weighted summation of the obstacle density, the number of dynamic objects, the rate of environmental change, and the proportion of passable areas, and then normalizing the summation result to a preset interval.

[0126] In this application, obstacle density is an indicator used to describe the density of obstacle distribution in an environment. Obstacle density can be defined as the number of obstacles per unit area or the proportion of space they occupy. Higher obstacle density means more limited space for robot passage, and a corresponding increase in obstacle avoidance difficulty. Obstacle density can be calculated using kernel density estimation methods or grid statistical methods, etc.

[0127] The number of dynamic objects refers to the number of obstacles in motion in the environment. The presence of dynamic objects increases the complexity of obstacle avoidance because the robot needs to predict their trajectories and adjust its obstacle avoidance strategy accordingly.

[0128] The number of dynamic objects can be identified and counted from sensor data in consecutive frames using a multi-target tracking algorithm.

[0129] As described in the previous embodiments, the rate of environmental change is an indicator used to describe the degree of drastic dynamic changes in the environment, and is calculated based on the difference norm of sensor readings between consecutive frames.

[0130] The passable area ratio refers to the proportion of the area in the environment where the robot can safely pass out to the total sensing area. A lower passable area ratio means that obstacles occupy more space, the robot's movement space is more restricted, and obstacle avoidance becomes more difficult.

[0131] The proportion of passable areas can be obtained by rasterizing the sensor data and calculating the ratio of the number of free grids to the total number of grids.

[0132] Weighted summation is a calculation method that linearly combines multiple indicators according to their respective weight coefficients. By setting reasonable weights for different indicators, the relative contribution of each indicator to environmental complexity can be reflected.

[0133] Environmental complexity value C The calculation can be performed using the following formula: ; Among them, obstacle density Number of dynamic objects Environmental change rate and the proportion of passable areas , , , and These are the corresponding weighting coefficients.

[0134] It should be noted that in the above formula, the proportion of passable area... With (1- The calculation is performed in the form of accessible areas because there is a negative correlation between the proportion of accessible areas and environmental complexity: the higher the proportion of accessible areas, the lower the environmental complexity; the lower the proportion of accessible areas, the higher the environmental complexity.

[0135] Normalization to a preset interval is the process of mapping the result of a weighted sum to a specified numerical range. The preset interval is usually set to [0,1], where 0 represents the lowest complexity and 1 represents the highest complexity. Normalization can be performed using the min-max normalization method.

[0136] This invention comprehensively considers multiple dimensions of factors, such as obstacle density, number of dynamic objects, environmental change rate, and proportion of passable areas, to generate a quantitative index that fully reflects the complexity of the environment. Compared to methods using only a single index, the environmental complexity assessment fused with multi-dimensional indicators is more accurate and comprehensive, providing a more reliable basis for subsequent adjustment of decision parameters.

[0137] Optionally, acquiring the robot's environmental data includes: The robot collects data about its surrounding environment using a set of sensors installed on it. Real-time monitoring of the rate of environmental change; wherein the rate of environmental change is obtained by calculating the difference norm of sensor readings between consecutive frames; The sampling frequency of the sensor array is dynamically adjusted according to the environmental change rate; wherein the sampling frequency is positively correlated with the environmental change rate.

[0138] In this application, the sensor group is a sensing unit composed of multiple sensors, used to collect various types of information about the robot's surrounding environment. The sensor group can include, but is not limited to, various types of sensors such as LiDAR, cameras, ultrasonic sensors, and infrared sensors.

[0139] LiDAR (Light Detection and Ranging) can accurately measure the distance between a robot and surrounding objects by emitting laser beams and receiving reflected signals, generating point cloud data or distance images of the environment. LiDAR has advantages such as high ranging accuracy and being unaffected by lighting conditions.

[0140] Cameras can capture visual image information of the environment, acquiring rich texture, color, and shape features. Through image processing and computer vision algorithms, obstacle types and moving objects can be identified from camera data.

[0141] Ultrasonic sensors can measure distance by emitting ultrasonic waves and receiving the echoes. They are particularly suitable for detecting transparent objects such as glass and can compensate for the shortcomings of lidar and cameras in detecting transparent obstacles.

[0142] By working together with multiple sensors, the sensor array can collect more comprehensive and accurate environmental data, improving the reliability and robustness of environmental perception.

[0143] Real-time monitoring refers to the process of continuously calculating and updating the rate of change of the environment during robot operation. By monitoring the rate of change of the environment in real time, the system can perceive the dynamic characteristics of the environment and provide a basis for subsequent adjustments to the sampling frequency.

[0144] In this embodiment, the environmental change rate is obtained by calculating the difference norm of sensor readings between consecutive frames. The specific calculation method is as described in the previous embodiment: ; in, Number of sensor data points For a moment t The i Sensor readings, For a moment t-1 The i Sensor readings.

[0145] Sampling frequency refers to the number of times a sensor array collects data per unit of time. Sampling frequency determines the temporal resolution of environmental perception: a higher sampling frequency can capture environmental changes more promptly, but it increases computational load and energy consumption; a lower sampling frequency can reduce computational load, but it may miss rapidly occurring environmental changes.

[0146] In this embodiment, the sampling frequency is positively correlated with the environmental change rate. Specifically, when the environmental change rate is high (e.g., there are fast-moving objects in the environment), the system automatically increases the sampling frequency to capture dynamic changes with higher time resolution, ensuring that no key environmental information is missed; when the environmental change rate is low, such as when the environment is relatively static, the system automatically reduces the sampling frequency to reduce computational load and energy consumption while ensuring the reliability of perception.

[0147] As a specific implementation method, the sampling frequency can be adjusted within a range of 5Hz to 50Hz. Lower sampling frequencies, such as 5Hz to 10Hz, are used in static environments, while higher sampling frequencies, such as 30Hz to 50Hz, are used in dynamic environments. The specific relationship between the sampling frequency and the rate of environmental change can be determined through linear mapping or piecewise functions.

[0148] The embodiments of the present invention can adaptively adjust the sampling frequency of the sensor according to the dynamic characteristics of the environment. In dynamic environments, the sampling frequency is increased to ensure timely detection of changes, while in static environments, the sampling frequency is decreased to reduce resource consumption, thereby achieving a good balance between real-time sensing and system resource utilization.

[0149] Optionally, a control signal is generated according to the obstacle avoidance action command, including: By using the high-level trajectory planning layer in the hierarchical control architecture, cubic spline interpolation is used to convert the path points corresponding to the obstacle avoidance action command into a smooth path. The bottom tracking control layer in the hierarchical control architecture uses a controller based on the proportional-integral-derivative algorithm to track the smooth path and output the control signal. When generating the control signal, a feedforward compensation term is introduced to offset the system inertial delay.

[0150] In this application, a hierarchical control architecture is a control system structure that decomposes control tasks into multiple layers for processing. In this embodiment, the hierarchical control architecture includes at least a high-level trajectory planning layer and a low-level tracking control layer. The high-level trajectory planning layer is responsible for generating smooth motion trajectories based on obstacle avoidance commands; the low-level tracking control layer is responsible for controlling the robot to accurately track the planned trajectory. Through the design of the hierarchical architecture, complex control tasks can be decomposed into relatively independent sub-tasks, improving the modularity and maintainability of the control system.

[0151] The feedforward compensation term is a control component introduced into the control signal to actively compensate for system characteristics. The robot's actuators have a certain inertia; when the control signal changes, the actuator's response is delayed, resulting in a lag between the actual and desired motion. The feedforward compensation term can adjust the control output in advance according to the changing trend of the control signal, offsetting the delay caused by system inertia and making the robot's response more timely.

[0152] The feedforward compensation term can be calculated using the differential form of the control signal: ; in, For feedforward compensation term, This is the feedforward gain coefficient. This is the derivative of the control signal with respect to time. The formula makes the feedforward compensation term proportional to the rate of change of the control signal: when the control signal changes rapidly, the feedforward compensation term is large, effectively offsetting inertial delay; when the control signal changes slowly, the feedforward compensation term is small, preventing additional disturbance to the system.

[0153] Smooth paths can be generated using cubic spline interpolation, a path smoothing method that can generate smooth curved trajectories based on given path points. Paths generated by cubic spline interpolation exhibit second-order continuity, meaning that the path's position, velocity, and acceleration change continuously, avoiding abrupt changes during motion and making the robot's movement more natural and smooth.

[0154] The mathematical form of cubic spline interpolation is a piecewise polynomial function: ; in, Let i be the spline function of the i-th segment. , , , For polynomial coefficients, Given the position of the i-th node, by solving the system of equations that satisfy the boundary and continuity conditions, the coefficients of the spline functions of each segment can be determined, thus obtaining the complete smooth path.

[0155] The proportional-integral-derivative (PID) algorithm is a classic feedback control algorithm that generates a control signal by calculating the proportional, integral, and derivative terms of the error. The PID algorithm is simple in structure, easy to tune parameters, and widely applicable, making it a widely used control method in industrial control.

[0156] In this embodiment, the underlying controller uses an improved proportional-integral-derivative algorithm for path tracking.

[0157] control signals u(t) The calculation formula is: ; in, This refers to the tracking error (the difference between the actual position and the desired position). For proportional gain, For integral gain, For differential gain, This is the second-order differential gain. Compared to the standard proportional-integral-differential algorithm, this improved algorithm adds a second-order differential term. This can improve the system's responsiveness to changes in acceleration, making control more precise.

[0158] The control parameters can be adaptively adjusted according to the robot's motion state: a more conservative parameter configuration is used during low-speed motion to ensure stability; and a more sensitive parameter configuration is used during high-speed motion to improve tracking accuracy.

[0159] This invention can convert high-level obstacle avoidance commands into smooth and accurate low-level control signals. Cubic spline interpolation ensures the smoothness of the motion trajectory, the improved proportional-integral-differential algorithm ensures the accuracy of trajectory tracking, and the feedforward compensation term offsets the effect of system inertial delay.

[0160] The present invention also provides an anomaly handling mechanism to improve the robustness and reliability of the robot's autonomous obstacle avoidance control system.

[0161] During the actual operation of a robot, sensors may generate abnormal data due to environmental interference, hardware failures, or other reasons. Sudden changes in the environment may also prevent the normal decision-making process from being effectively executed.

[0162] To address these anomalies, this invention constructs a graded fault detection and recovery strategy. The anomaly detection module performs multi-dimensional monitoring of environmental data collected by sensors to identify potential anomalies.

[0163] The multi-dimensional monitoring covers three aspects: data consistency check, temporal continuity verification, and spatial rationality check.

[0164] Data consistency checks cross-validate data collected by multiple sensors to detect measurement conflicts between different sensors. For example, if a lidar detects a nearby obstacle in a certain direction while an ultrasonic sensor does not detect an obstacle in the same direction, it may indicate an anomaly in one of the sensors. Timing continuity verification monitors the update frequency and timestamps of sensor data to verify the timeliness of updates. When an abnormal delay or interruption occurs in the data update of a sensor, the system can promptly detect it and take appropriate measures.

[0165] The spatial rationality check verifies whether the sensor measurements conform to physical laws and spatial constraints. For example, the rate of change of distance to obstacles should not exceed the physical maximum possible speed, and the distance measurement should not be negative or exceed the sensor's range.

[0166] When the detected abnormal indicators exceed the preset threshold, the system determines that the current environmental data quality does not meet the requirements for normal decision-making and triggers the corresponding fault recovery strategy.

[0167] Depending on the severity of the fault, this invention employs a graded recovery strategy to ensure that the system maintains its corresponding functions under different levels of anomalies.

[0168] When a single sensor malfunctions or detects abnormal data, the system performs a first-level recovery, which involves adjusting the weight allocation during the multi-sensor fusion process, reducing the contribution weight of data from the faulty sensor, and increasing the contribution weight of data from the normal sensor. In this way, the system can continue to use data from other sensors for environmental perception and maintain basic obstacle avoidance functions even when a single sensor malfunctions.

[0169] When multiple sensors malfunction simultaneously, making it impossible to obtain sufficiently reliable environmental data at the current moment, the system initiates a second-level recovery mode, namely the data prediction mode. This mode uses an autoregressive model to predict the current environmental state based on historical data. The autoregressive model is as follows: ; Among them, For the current data used in the forecast, For historical data, i These are the autoregressive coefficients. This represents the prediction error.

[0170] The autoregressive model system can quickly fill in missing sensor data and maintain the continuity of obstacle avoidance control.

[0171] When environmental data quality consistently falls below a preset threshold or a system-level failure occurs, the system automatically switches to the third level of recovery, namely the rule-based rollback mode. In rollback mode, the system uses historical reliable data and preset safety rules to generate obstacle avoidance actions, rather than relying on real-time decision model output. This rollback mode uses an artificial potential field method for obstacle avoidance, with an attractive potential guiding the robot to move towards the target position and a repulsive potential ensuring that the robot maintains a safe distance from known obstacles. Although the obstacle avoidance performance of rollback mode may not be as good as the decision model output in normal mode, it can ensure the basic safety of the robot in abnormal situations.

[0172] In addition, the system records all detected abnormal events, including the type of abnormality, the time of occurrence, the duration, the recovery measures taken and their effects. These records can be used for subsequent system analysis and model improvement, helping developers identify system weaknesses and optimize sensor configurations and algorithm parameters.

[0173] Through the above-mentioned anomaly handling mechanism, the present invention can maintain basic obstacle avoidance function in abnormal situations such as sensor failure or sudden environmental changes, thereby improving the fault tolerance and operational reliability of the system. Compared with the prior art which lacks anomaly handling capabilities, the present invention can better cope with various abnormal situations that may be encountered in actual operation and ensure the safe operation of the robot.

[0174] Optionally, based on the above embodiments, the present invention also provides a behavior management method for coordinating obstacle avoidance control strategies of a robot under different task modes. In a home environment, pet robots need to perform various types of tasks, and different tasks have different requirements for obstacle avoidance control. The present invention uses a behavior management module to adapt and adjust decision parameters under different task modes.

[0175] This invention defines multiple task modes to adapt to different application scenarios. In follow mode, the robot follows a specific target, requiring it to maintain an appropriate distance from the target while avoiding obstacles along the way. The optimization objective of follow mode is to minimize the distance error and directional deviation from the target while maintaining smooth motion. Its cost function can be expressed as: ; in, To account for the distance error relative to the target being followed, For directional deviation, As an index of motion smoothness, , , These are the corresponding weighting coefficients.

[0176] In exploration mode, the robot autonomously explores the environment, aiming to cover as much area as possible while avoiding obstacles. Exploration mode emphasizes environmental coverage and information acquisition, and its reward function can be expressed as: ; in For the area of ​​the newly explored region, For information gain, For safety risks, α , β , γ For the corresponding coefficient.

[0177] In rest mode, the robot is in standby or low-power state, requiring minimization of energy consumption and collision risk. The objective function for rest mode can be expressed as: ; in, For power consumption, To mitigate collision risk, These are the weighting coefficients.

[0178] In some application scenarios, the system can automatically identify the current task mode based on environmental characteristics and user behavior and automatically switch modes. For example, when a specific target is detected within the perception range, the system can automatically switch to follow mode, and when no interaction needs are detected for a long time, the system can automatically switch to rest mode.

[0179] Through the above-mentioned behavior management method, the present invention can adjust the obstacle avoidance control strategy according to different task requirements, so that the robot can maintain good obstacle avoidance performance when performing different tasks. The behavior management module works in conjunction with the aforementioned parameter adaptive adjustment mechanism to achieve dual adaptation at the task level and the environment level, further improving the intelligence level of the robot's autonomous obstacle avoidance control.

[0180] Through the above-mentioned anomaly handling mechanism and behavior management extended implementation method, the robot autonomous obstacle avoidance control method provided by the present invention has a more complete system architecture. The anomaly handling mechanism ensures the safe operation of the system under various abnormal conditions, and the behavior management mechanism realizes the flexible adaptation of obstacle avoidance strategies under different task modes. These extended functions, combined with the aforementioned core technical solutions, enable the present invention to exhibit good adaptability, reliability and intelligence in complex and ever-changing real-world application environments.

[0181] The robot provided by the present invention is described below. The robot described below and the robot autonomous obstacle avoidance control method described above can be referred to in correspondence.

[0182] Figure 2 This is a schematic diagram of the robot structure provided by the present invention, such as... Figure 2 As shown, it includes: Sensor group 210 is used to collect environmental data around the robot; Memory 220 is used to store computer programs; Processor 230 is configured to execute the computer program to perform the following steps: The system acquires environmental data collected by the sensor array, processes the environmental data, and generates an environmental state description. The environmental complexity value is calculated based on the environmental state description. The exploration rate, learning rate, and reward function weights of the decision model are adjusted in real time according to the environmental complexity value. The exploration rate is determined based on the environmental complexity value and real-time performance feedback data. The learning rate is determined based on the environmental complexity value and the environmental change rate. The safety weight and efficiency weight in the reward function are dynamically scaled based on the environmental complexity value and obstacle type. The environmental state description is input into the adjusted decision model, which outputs obstacle avoidance action instructions and generates control signals based on the obstacle avoidance action instructions. Actuator 240 is used to respond to the control signal to drive the robot to move.

[0183] In this embodiment of the invention, the exploration rate, learning rate, and reward function weights of the decision model can be adjusted in real time according to the environmental complexity value, enabling the robot's obstacle avoidance strategy to adaptively match the current environmental characteristics. When the environment is complex, the exploration rate is increased to promote the discovery of new strategies, the learning rate is reduced to ensure learning stability, and safety weights are increased to prioritize obstacle avoidance safety. When the environment is simple, the system utilizes existing experience by reducing the exploration rate, improves learning efficiency by maintaining an appropriate learning rate, and enhances movement efficiency by increasing efficiency weights. This dynamic adjustment mechanism overcomes the shortcomings of fixed-parameter models in existing technologies that cannot adapt to environmental changes, achieving adaptive obstacle avoidance control without human intervention, and significantly improving the robot's movement efficiency while ensuring safety.

[0184] Figure 3 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 3 As shown, the electronic device may include a processor 310, a communications interface 320, a memory 330, and a communication bus 340. The processor 310, communications interface 320, and memory 330 communicate with each other via the communication bus 340. The processor 310 can call logical instructions from the memory 330 to execute a robot autonomous obstacle avoidance control method. This method includes: acquiring environmental data of the robot and processing the environmental data to generate an environmental state description. The environmental complexity value is calculated based on the environmental state description. The exploration rate, learning rate, and reward function weights of the decision model are adjusted in real time according to the environmental complexity value. The exploration rate is determined based on the environmental complexity value and real-time performance feedback data. The learning rate is determined based on the environmental complexity value and the environmental change rate. The safety weight and efficiency weight in the reward function are dynamically scaled based on the environmental complexity value and obstacle type. The environmental state description is input into the adjusted decision model, which outputs obstacle avoidance action commands. Control signals are generated based on the obstacle avoidance action commands to control the robot's movement.

[0185] Furthermore, the logical instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, 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 computer 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 the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0186] On the other hand, the present invention also provides a computer program product, the computer program product including a computer program, the computer program being able to be stored on a non-transitory computer-readable storage medium, the computer program being executed by a processor, the computer being able to execute the robot autonomous obstacle avoidance control method provided by the above methods, the method including: acquiring environmental data of the robot, and processing the environmental data to generate an environmental state description; The environmental complexity value is calculated based on the environmental state description. The exploration rate, learning rate, and reward function weights of the decision model are adjusted in real time according to the environmental complexity value. The exploration rate is determined based on the environmental complexity value and real-time performance feedback data. The learning rate is determined based on the environmental complexity value and the environmental change rate. The safety weight and efficiency weight in the reward function are dynamically scaled based on the environmental complexity value and obstacle type. The environmental state description is input into the adjusted decision model, which outputs obstacle avoidance action commands. Control signals are generated based on the obstacle avoidance action commands to control the robot's movement.

[0187] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the robot autonomous obstacle avoidance control method provided by the above methods, the method comprising: acquiring environmental data of the robot, and processing the environmental data to generate an environmental state description; The environmental complexity value is calculated based on the environmental state description. The exploration rate, learning rate, and reward function weights of the decision model are adjusted in real time according to the environmental complexity value. The exploration rate is determined based on the environmental complexity value and real-time performance feedback data. The learning rate is determined based on the environmental complexity value and the environmental change rate. The safety weight and efficiency weight in the reward function are dynamically scaled based on the environmental complexity value and obstacle type. The environmental state description is input into the adjusted decision model, which outputs obstacle avoidance action commands. Control signals are generated based on the obstacle avoidance action commands to control the robot's movement.

[0188] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0189] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0190] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A robot autonomous obstacle avoidance control method, characterized by, include: Acquire environmental data of the robot and process the environmental data to generate an environmental state description; The environmental complexity value is calculated based on the environmental state description. The exploration rate, learning rate, and reward function weights of the decision model are adjusted in real time according to the environmental complexity value. The exploration rate is determined based on the environmental complexity value and real-time performance feedback data. The learning rate is determined based on the environmental complexity value and the environmental change rate. The safety weight and efficiency weight in the reward function are dynamically scaled based on the environmental complexity value and obstacle type. The environmental state description is input into the adjusted decision model, which outputs obstacle avoidance action commands. Control signals are generated based on the obstacle avoidance action commands to control the robot's movement. 2.The robot autonomous obstacle avoidance control method of claim 1, wherein, The method for determining the exploration rate includes: The environmental complexity value and real-time performance feedback data are obtained; wherein, the real-time performance feedback data includes the number of consecutive successful obstacle avoidances and near-collision events of the robot; Based on the environmental complexity value, a basic exploration rate is calculated using a nonlinear mapping function; wherein, the basic exploration rate increases with the increase of the environmental complexity value; The exploration rate adjustment is calculated based on the real-time performance feedback data, and the exploration rate is obtained by summing the base exploration rate and the exploration rate adjustment.

3. The robot autonomous obstacle avoidance control method according to claim 2, characterized in that, The nonlinear mapping function is the Sigmoid function; the basic exploration rate is in the first exploration rate interval when the environmental complexity value is less than the first complexity threshold, and in the second exploration rate interval when the environmental complexity value is greater than the second complexity threshold, and the value of the second exploration rate interval is greater than the value of the first exploration rate interval; When the number of consecutive successful obstacle avoidances reaches a preset threshold, the exploration rate adjustment amount is increased; when a near-collision event is detected, the exploration rate adjustment amount is decreased.

4. The robot autonomous obstacle avoidance control method according to claim 1, characterized in that, The method for determining the learning rate includes: Based on the environmental complexity value, the decay factor is calculated using the exponential decay function; Based on the environmental change rate, the stability factor is calculated using the stability compensation function. The learning rate is obtained based on the base learning rate, the decay factor, and the stability factor.

5. The robot autonomous obstacle avoidance control method according to claim 4, characterized in that, The method for determining the learning rate specifically includes: The environmental change rate is obtained; wherein the environmental change rate is calculated based on the difference norm of sensor readings between consecutive frames; The step of calculating the decay factor based on the environmental complexity value using an exponential decay function includes: calculating the decay factor so that the decay factor decreases as the environmental complexity value increases; The step of calculating the stability factor based on the environmental change rate using a stability compensation function includes: calculating the stability factor so that the stability factor decreases as the environmental change rate increases.

6. The robot autonomous obstacle avoidance control method according to claim 1, characterized in that, The method for determining the safety weight and efficiency weight in the reward function weights includes: The complexity scaling factor is determined based on the aforementioned environment complexity value; Identify the types of obstacles in the environmental state description and determine the corresponding type coefficients; wherein, the type coefficients for dynamic obstacles are greater than those for static obstacles; The security weights are obtained by positively scaling the security base weights based on the complexity scaling factor and the type coefficient. Based on the complexity scaling factor, the efficiency base weight is inversely scaled to obtain the efficiency weight. The reward function of the decision model is the product of the safety weight and the safety reward, the product of the efficiency weight and the efficiency reward, and the weighted sum of the target reward.

7. The robot autonomous obstacle avoidance control method according to claim 1, characterized in that, The environmental complexity value is calculated based on the environmental state description, including: Based on the environmental state description, the obstacle density, number of dynamic objects, environmental change rate, and proportion of passable area are calculated. The environmental complexity value is obtained by weighted summation of the obstacle density, the number of dynamic objects, the rate of environmental change, and the proportion of passable areas, and then normalizing the summation result to a preset interval.

8. The robot autonomous obstacle avoidance control method according to claim 1, characterized in that, The acquisition of the robot's environmental data includes: The robot collects data about its surrounding environment using a set of sensors installed on it. Real-time monitoring of the rate of environmental change; wherein the rate of environmental change is obtained by calculating the difference norm of sensor readings between consecutive frames; The sampling frequency of the sensor array is dynamically adjusted according to the environmental change rate; wherein the sampling frequency is positively correlated with the environmental change rate.

9. The robot autonomous obstacle avoidance control method according to claim 1, characterized in that, Control signals are generated based on the obstacle avoidance action commands, including: By using the high-level trajectory planning layer in the hierarchical control architecture, cubic spline interpolation is used to convert the path points corresponding to the obstacle avoidance action command into a smooth path. The bottom tracking control layer in the hierarchical control architecture uses a controller based on the proportional-integral-derivative algorithm to track the smooth path and output the control signal. When generating the control signal, a feedforward compensation term is introduced to offset the system inertial delay.

10. A robot, characterized in that, include: A sensor array is used to collect environmental data around the robot; Memory, used to store computer programs; A processor is configured to execute the computer program to perform the following steps: The system acquires environmental data collected by the sensor array, processes the environmental data, and generates an environmental state description. The environmental complexity value is calculated based on the environmental state description. The exploration rate, learning rate, and reward function weights of the decision model are adjusted in real time according to the environmental complexity value. The exploration rate is determined based on the environmental complexity value and real-time performance feedback data. The learning rate is determined based on the environmental complexity value and the environmental change rate. The safety weight and efficiency weight in the reward function are dynamically scaled based on the environmental complexity value and obstacle type. The environmental state description is input into the adjusted decision model, which outputs obstacle avoidance action instructions and generates control signals based on the obstacle avoidance action instructions. An actuator is used to respond to the control signal to drive the robot to move.

11. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the robot autonomous obstacle avoidance control method as described in any one of claims 1 to 9.

12. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the robot autonomous obstacle avoidance control method as described in any one of claims 1 to 9.

Images