Uncertainty-aware multi-joint robot arm path planning system, method, electronic device, storage medium

Abstract

The application discloses a multi-joint robot path planning system and method based on uncertainty perception, an electronic device and a storage medium. The multi-joint robot path planning system based on uncertainty perception comprises a perception module configured to obtain obstacle information corresponding to a multi-joint robot; an obstacle model module configured to construct obstacles into dynamic obstacle occupancy with uncertainty boundaries; a candidate path module configured to obtain a candidate path of the multi-joint robot; a robot model module configured to generate a sweep volume of the multi-joint robot; a risk budget module configured to set a risk budget and then distribute the risk budget to edges of the candidate path; a collision constraint module configured to set collision constraints; and a global planning module configured to construct a plurality of to-be-optimized paths according to edges of the candidate path satisfying the collision constraints, and then obtain a safe path from the to-be-optimized paths through comprehensive cost optimization. The application can plan a path of the multi-joint robot in a dynamic obstacle scene.

Description

Technical Field

[0001] This application belongs to the technical field of robotic arms, specifically relating to a path planning system, method, electronic device, and storage medium for a multi-joint robotic arm based on uncertainty perception. Background Technology

[0002] The structure of a multi-joint robotic arm (also known as a multi-axis robotic arm or multi-degree-of-freedom robotic arm) can typically be modeled as consisting of a fixed base / control cabinet, multiple joints, one or more end effectors, and multiple links connecting the fixed base / control cabinet, multiple joints, and end effectors. It is widely used in production scenarios such as welding, loading / unloading, palletizing, sorting, assembly, grinding, semiconductor process transfer, and human-robot collaboration. Path planning for multi-joint robotic arms requires comprehensive consideration of joint limits, speed limits, acceleration limits, torque limits, end effector posture constraints, obstacle avoidance, production line cycle time, and safety standards to obtain a path that meets the desired objectives. With the increase in flexible manufacturing and open workstations, the obstacles faced by multi-joint robotic arms during path planning are no longer just static, geometrically defined objects, but also dynamic obstacles formed by personnel, AGVs, pallets, bins, temporary tooling, and other robotic arms. The position, speed, contour, and sensor measurements of the multi-joint robotic arm relative to the dynamic obstacle objects will become uncertain due to measurement noise or changes, resulting in changes in the model boundaries of the dynamic obstacle objects.

[0003] In existing technologies, the main technical routes for path planning of multi-joint robotic arms include RRT and RRT-3. Sampling planning systems such as Bi-RRT and HybridRRT are used, along with hybrid architectures combining global planning and local real-time obstacle avoidance, joint space graph search, roadmap and multi-planner switching, multi-objective trajectory optimization (time / energy consumption / acceleration / jerkiness), digital twins, reinforcement learning, and dynamic multi-criteria decision-making. These technological approaches have propelled multi-joint robotic arms from simply finding collision-free paths to generating executable, optimizable, and online-updable objectives.

[0004] In existing technologies, RRT-type solutions typically improve search efficiency and path smoothness through target bias, dynamic step size, parent node reselection, cost reconnection, or spline smoothing. Local obstacle avoidance solutions utilize artificial potential fields, MPC, dynamic systems, or local replanning of response obstacles. Multi-objective optimization solutions optimize metrics such as time, energy consumption, impact, and smoothness using B-splines, genetic algorithms, NSGA-II, or reinforcement learning. While existing solutions are effective for known obstacles in a given model, they suffer from drawbacks due to sensor noise, calibration errors, occlusion, time delay, and human movement. (i) Existing technologies treat obstacle boundaries as defined boundaries or only use fixed expansion safety distances without distinguishing between high-confidence and low-confidence measurements, resulting in overly conservative low-risk scenarios and insufficient safety margins in high-risk scenarios. (ii) The feasibility of paths in the prior art is often judged by "geometric non-intersection" or "distance greater than the threshold", lacking information on collision probability, risk budget, confidence boundary, etc., and cannot provide auditable safety proof; (iii) Information may be disconnected between front-end planning and back-end control. The safe path passed in the planning stage may fail in the execution stage due to tracking errors, communication delays, changes in obstacle estimation, etc., causing the controller's current action to no longer be adapted to the path target. (iv) When the existing technology encounters situations such as partial replanning or emergency stop, it fails to record information such as path certificate, cause of failure, rollback trajectory, and log archiving to form a closed loop of path data; (v) The existing technologies, such as RRT parameter fine-tuning and artificial potential field correction, which continue to make small changes within a limited area, lack stability. Summary of the Invention

[0005] To address one or more of the aforementioned problems, the first aspect of this application provides a multi-joint robotic arm path planning system based on uncertainty perception, the second aspect provides a multi-joint robotic arm path planning method based on uncertainty perception, the third aspect provides an electronic device for implementing the second aspect, and the fourth aspect provides a storage medium for implementing the second aspect, thereby avoiding overly conservative or insufficient safety margins in path planning, adapting to changes in dynamic obstacle objects, and improving the stability of path planning.

[0006] The technical solution of this application is as follows.

[0007] In a first aspect, this application discloses a path planning system for a multi-joint robotic arm based on uncertainty perception, comprising: The perception module is used to acquire obstacle information for the multi-joint robotic arm; The obstacle model module is used to construct obstacles into dynamic obstacle occupants with uncertain boundaries based on obstacle information; The candidate path module is used to generate candidate nodes in the joint space of the multi-joint robot arm, and then remove candidate nodes that violate joint limits, self-collision, and static safety envelope to obtain candidate paths for the multi-joint robot arm. The robotic arm model module is used to construct a multi-joint robotic arm as a robotic arm occupant with defined boundaries, and then generate the sweep body of the multi-joint robotic arm based on the robotic arm occupant and candidate paths. The risk budget module is used to set the global allowable collision probability upper limit between the swept body and the dynamically occupied obstacle body as the risk budget, and then allocate the risk budget to the edges of the candidate path as the corresponding local allowable collision probability upper limit. The collision constraint module is used to set the collision probability between the swept body and the dynamically occupied obstacle body as the upper limit of the local allowed collision probability assigned to the edge of the corresponding candidate path as a collision constraint. The global planning module is used to construct multiple paths to be optimized based on the edges of candidate paths that satisfy collision constraints, and then obtain safe paths from the paths to be optimized through comprehensive cost optimization.

[0008] One implementation of the first aspect further includes a security certificate module for generating security certificates; Safety certificates are used to record data related to the dynamic occupancy of obstacles, candidate paths, swept volumes, risk budget allocation, collision constraints, and safe paths in a structured manner. And / or, it also includes a local safety filtering module, which is used to make real-time corrections to the multi-joint robotic arm without changing the safety path, so that the multi-joint robotic arm meets collision constraints when it moves. And / or, it also includes a degraded execution module, used to set the safety level when the multi-joint robotic arm executes the corresponding safety path, and to control the multi-joint robotic arm to work in the corresponding action mode according to the safety level.

[0009] Preferably, the local safety filtering module corrects collision constraints in real time by converting them into control constraints for the multi-joint robotic arm.

[0010] In one implementation of the first aspect, the obstacle dynamic occupancy volume is constructed using one or more of the following methods: ellipsoid, sphere, capsule, bounding box, occupancy grid, signed distance field, Gaussian object, occupancy octree, interval envelope, fuzzy set, reachability set, and neural network prediction.

[0011] In one embodiment of the first aspect, the robotic arm occupying body is one or more of the following: a capsule, a sphere, a bounding box, and a mesh; The swept body is formed by the robotic arm occupying the body through the candidate path.

[0012] In one implementation of the first aspect, the risk budget is allocated using a dynamic allocation mechanism.

[0013] In one implementation of the first aspect, the edges of candidate paths that satisfy collision constraints are added to the global path graph, and a global planning algorithm is used to obtain the path to be optimized. Comprehensive cost optimization includes one of the following: weighted summation, Pareto selection, multi-objective optimization, and hierarchical optimization.

[0014] Secondly, this application provides a path planning method for a multi-joint robotic arm based on uncertainty perception, comprising: Obtain obstacle information corresponding to the multi-joint robotic arm; Based on obstacle information, obstacles are constructed as dynamic obstacle occupants with uncertain boundaries; Candidate nodes are generated in the joint space of the multi-joint robotic arm, and then candidate nodes that violate joint limits, self-collision, and static safety envelope are eliminated to obtain candidate paths for the multi-joint robotic arm. After constructing the multi-joint robotic arm as a robotic arm occupying body with defined boundaries, the sweeping body of the multi-joint robotic arm is generated based on the robotic arm occupying body and candidate paths. Set the global allowed collision probability cap between the swept body and the dynamically occupied obstacle body as the risk budget, and then allocate the risk budget to the edges of the candidate path as the corresponding local allowed collision probability cap. The collision constraint is set to ensure that the collision probability between the swept body and the dynamically occupied obstacle is no greater than the upper limit of the local allowed collision probability assigned to the edge of the corresponding candidate path. Multiple paths to be optimized are constructed based on the edges of candidate paths that satisfy collision constraints. Then, a safe path is obtained from the paths to be optimized through comprehensive cost optimization.

[0015] Thirdly, an electronic device according to this application includes a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the uncertainty-aware multi-joint robotic arm path planning method of the second aspect.

[0016] Fourthly, one storage medium of this application is a computer-readable storage medium storing computer instructions for enabling a computer to implement the uncertainty-aware multi-joint robotic arm path planning method of the second aspect.

[0017] Compared with the prior art, the advantages of this application are as follows: Instead of relying solely on deterministic geometric distances to determine path safety, this approach provides quantifiable safety credentials for each path and its edges. It reduces unnecessary boundary occupancy and improves traffic efficiency when the uncertainty in measuring dynamic obstacles is low; and expands boundary occupancy to avoid insufficient safety margins when the uncertainty in measuring dynamic obstacles is high or dynamic risks increase. This enhances the safety, interpretability, and engineering deployability of planned paths in human-machine collaboration, open production lines, dynamic sorting, and high-value automated equipment. Attached Figure Description

[0018] Figure 1This is a schematic diagram of the structural framework of one of the multi-joint robotic arm path planning systems based on uncertainty perception in this application.

[0019] Figure 2 This is a flowchart illustrating one of the path planning methods for a multi-joint robotic arm based on uncertainty perception in this application.

[0020] Figure 3 This is a structural framework diagram of an electronic device according to this application. Detailed Implementation

[0021] Referring to the illustrations, the principles of this application are illustrated by way of example implementation in a suitable operating environment. The following description is based on the illustrative specific embodiments of this application and should not be construed as limiting other specific embodiments not detailed herein, such as corresponding adjustments to the order of steps of the method of this application based on the principles of the technical concept.

[0022] The multi-joint robotic arms applicable to this application include, but are not limited to, those primarily based on six-joint robotic arms, but which can be extended to seven-axis redundant robotic arms, SCARA robots, collaborative robots, mobile robotic arms, dual-arm robots, and industrial robots with external axes.

[0023] like Figure 1 As shown in the figure, this application provides a multi-joint robotic arm path planning system based on uncertainty perception, including a perception module, an obstacle model module, a candidate path module, a robotic arm model module, a risk budget module, a collision constraint module, a global planning module, a safety certificate module, a local safety filtering module, and a degraded execution module.

[0024] The perception module is used to acquire obstacle information for the multi-joint robotic arm.

[0025] The obstacle model module is used to construct obstacles into dynamic occupants with uncertain boundaries based on obstacle information.

[0026] Optionally, in this embodiment, the perception module can acquire information about one or more obstacles in the environment surrounding the multi-joint robotic arm by connecting to one or more information acquisition devices, including but not limited to monocular / dual-lens cameras, structured light cameras, TOF cameras, RGB-D cameras, 2D / 3D LiDAR, millimeter-wave radar, ultrasonic sensors, force / torque sensors, safety mats, safety doors, and external motion capture systems. The obstacles can be static or dynamic. In this embodiment, obstacles are preferably treated as dynamic objects. The specific information acquisition device used only needs to be adjusted according to the actual environment where the multi-joint robotic arm is located.

[0027] Optionally in this embodiment, after acquiring one or more obstacle information, the perception module calibrates the intrinsic / extrinsic parameters of the information acquisition device and the base coordinate system of the multi-joint robotic arm. Then, based on the acquisition time of the obstacle information and the control cycle of the multi-joint robotic arm, it performs time synchronization and delay compensation on the obstacle information and timestamps the obstacle information. For dynamic objects that are constantly changing, such as personnel, material bins, tooling, mobile carts, and other robots, their outlines can be obtained through clustering, target detection, semantic segmentation, tracking algorithms, etc.

[0028] The obstacle dynamic occupancy volume can be constructed using one or more of the following methods: ellipsoid, sphere, capsule, bounding box, occupancy grid, signed distance field, Gaussian object, occupancy octree, interval envelope, fuzzy set, reachability set, and neural network prediction. It is only necessary that the given obstacle dynamic occupancy volume has an uncertain boundary that can be used to calculate the collision probability.

[0029] In this embodiment, the obstacle information that the sensing module needs to acquire may include, but is not limited to, the shape outline, category, and spatial pose changes of the obstacle, and may be selected according to the needs of the dynamic occupant of the corresponding obstacle.

[0030] Optionally, in this embodiment, when the obstacle model module selects a capsule as the model for the dynamic occupant of the obstacle, the perception module also processes the obstacle information in advance, recording all the data to be used by the obstacle model module as structured state data. This structured state data includes, but is not limited to, the geometric shape / contour template of the obstacle, the average position / pose of the obstacle in the robot arm's base coordinate system, the covariance of the obstacle's position / attitude / contour boundary, the predicted velocity of the obstacle, the predicted acceleration of the obstacle, the category of the obstacle, the category confidence of the obstacle, the current information validity period (which can be manually set), and the model version number of the obstacle's dynamic occupant in the current path planning cycle. The covariance of the obstacle's position / attitude / contour boundary can be further obtained by superimposing multiple measurement error sources, including, but not limited to, sensor ranging noise. The covariance calculation can be performed by accumulating and conservatively synthesizing multiple errors, such as point cloud fitting residuals, hand-eye / extrinsic parameter calibration errors, target tracking residuals, boundary uncertainties caused by occlusion ratios, and prediction errors caused by communication and control delays. Alternatively, extended Kalman filtering, unscented Kalman filtering, particle filtering, and other methods can be used to output the predicted value of the superimposed error sources as the covariance result. The obstacle model module, based on the capsule constructed from structured state data, can be based on the covariance by multiplying the covariance confidence coefficient (which can be set manually or determined based on risk budget) by the square root of the largest eigenvalue of the covariance, adding the basic safety distance (which can be set manually), adding the measurement delay multiplied by the actual speed, and adding the calibration error upper limit to calculate the expansion radius of the capsule, thus forming the uncertainty boundary of the capsule. This gives the boundary of the capsule a probabilistic meaning and a validity period meaning.

[0031] In other implementations, when the obstacle model module selects a model with probabilistic boundaries, such as an occupied grid (a probabilistic occupied volume), for example, for an ellipsoid or Gaussian object, the corresponding uncertainty boundary can be determined based on the Mahalanobis distance threshold. The perception module also performs matching processing on the obstacle information in advance, which facilitates the rapid determination of the collision probability in the future.

[0032] In this embodiment, when the measurement noise of the obstacle decreases or the degree of spatial change decreases, the confidence level of whether the obstacle information obtained by the perception module is accurate increases. At this time, the uncertainty boundary of the obstacle's dynamic occupancy body shrinks and narrows, and the feasible space of path planning also increases, thus avoiding being too conservative. Conversely, the confidence level of the obstacle information decreases, which can make the uncertainty boundary expand relatively, thus avoiding insufficient safety margin. The model boundary of the obstacle is no longer single and definite, thus adapting to the path planning requirements in dynamic object scenarios.

[0033] The candidate path module is used to generate candidate nodes in the joint space of a multi-joint robotic arm, and then remove candidate nodes that violate joint limits, self-collision, and static safety envelope to obtain candidate paths for the multi-joint robotic arm.

[0034] Optionally in this embodiment, based on the current joint configuration and the target joint configuration of the multi-joint robotic arm, all available joint space nodes in the joint space from the current joint configuration to the target joint configuration are selected as candidate nodes. After the candidate nodes are eliminated, the remaining parts are connected by multiple segments to form candidate paths for each joint, with each segment serving as an edge of a candidate path. Optionally in this embodiment, when selecting all available joint space nodes, the forward kinematics model of the multi-joint robotic arm needs to be solved simultaneously to ensure that the available joint nodes are reachable in spatial position for each joint.

[0035] The robotic arm model module is used to construct a multi-joint robotic arm as a robotic arm occupant with defined boundaries, and then generate the sweep body of the multi-joint robotic arm based on the robotic arm occupant and candidate paths.

[0036] The robotic arm occupying body can be one or more of the following: capsule, sphere, bounding box, and mesh; the sweeping body is formed by the robotic arm occupying body through the candidate path.

[0037] In this embodiment, each link of the multi-joint robotic arm can be represented by one or more capsules with defined boundaries, the joints can be represented by spheres, and the end effector can be approximated by bounding boxes, capsules, or meshes. These are combined to form the model basis of the robotic arm's occupying body. Then, based on the DH parameters of the multi-joint robotic arm, the link dimensions, joint offsets, the center point of the end effector, and the safety envelope, the forward kinematics model of the multi-joint robotic arm from joint space to Cartesian space is solved to obtain the specific configuration of the robotic arm's occupying body. For example, for the first link, the endpoint of the capsule's centerline and the link's safety radius are calculated using forward kinematics. Then, the capsule shape boundary of the robotic arm's occupying body in the link portion is determined. The shape boundary of the joints and end effectors can be obtained by using the corresponding shape calculation methods of spheres, capsules, bounding boxes, or mesh envelopes, respectively. These are then combined with the link portion to obtain the entire robotic arm's occupying body.

[0038] Optionally in this embodiment, the way the robotic arm occupier generates the sweep body through the candidate path is as follows: when passing the edge of each candidate path, a number of joint configuration verification configurations are generated according to joint interpolation, trapezoidal / quintic polynomial time parameterization or dynamic feasible trajectory, and then the robotic arm occupiers corresponding to all verification configurations are collected as the sweep body. Optionally, in this embodiment, when a higher level of safety is required, the generated sweep volume can employ an adaptive subdivision approach: when the end displacement, joint angle change, or minimum distance gradient of adjacent joint configurations exceeds a set threshold, additional verification points for the generated joint configurations are added. To avoid missing collision constraints between adjacent verification points, an adaptive subdivision rule can be adopted: when adjacent joint nodes exceed joint thresholds, end displacements exceed Cartesian thresholds, the minimum distance change rate of adjacent joint configurations exceeds a threshold, or the predicted covariance of obstacles increases rapidly within a certain time period, further subdivision is performed within those adjacent joint nodes until the adjacent joint nodes of the sampling interval no longer meet the conditions of adjacent joint nodes exceeding joint thresholds, end displacements exceeding Cartesian thresholds, the minimum distance change rate of adjacent joint configurations exceeding a threshold, or the predicted covariance of obstacles increasing rapidly within a certain time period. In high-safety-level scenarios, the upper limit of the link endpoint velocity can also be multiplied by the time step and then divided by two to obtain the additional sweep margin of the link to cover discretization errors. In other embodiments, the sweep volume can be implemented using discrete capsule sequences, adjacent capsule convex hulls, bounding box trees, SDF voxels, or conservative Minkowski envelopes.

[0039] The risk budget module is used to set the global allowed collision probability upper limit between the swept body and the dynamically occupied obstacle body as the risk budget, and then allocate the risk budget to the edges of the candidate path as the corresponding local allowed collision probability upper limit.

[0040] Optionally in this embodiment, the risk budget module determines the global allowable collision probability upper limit based on the workstation safety mode. For example, in unmanned enclosed workstations, collaborative speed-limited workstations, and workstations with frequent personnel entry, different levels of total risk and minimum confidence requirements can be set respectively. The global allowable collision probability upper limit can be issued by the user, safety engineering parameters, or host computer, or it can be automatically selected according to the process risk level.

[0041] Optionally, in this embodiment, the risk budget can be allocated not only to the edges of the candidate path, but also to obstacles and time slices. The allocation of the risk budget can be uniform when the information on wide channels and obstacles is highly certain, and adaptive allocation can be used in narrow channels or areas with high uncertainty of obstacle information. A higher budget is allocated to the edges that are inevitably narrow channel candidate paths, while keeping the sum of the upper limits of the local allowed collision probabilities on all edges from not exceeding the upper limit of the global allowed collision probability. That is, the sum of the upper limits of the probability of occurrence of each local collision event does not exceed the risk budget. This condition can be used as conservative evidence for the occurrence of a collision event.

[0042] Optionally, in this embodiment, the risk budget allocation adopts a dynamic allocation mechanism. The risk budget is decomposed into path-level, edge-level, obstacle-level, link-level, and time-slice-level budgets to form a risk budget table. The initial allocation can be uniform, or the allocation weight can be set according to obstacle type, personnel accessibility, workstation safety level, local passage width, and uncertainty. In this embodiment, an optional dynamic allocation mechanism is adaptive allocation. The adaptive allocation method can comprehensively consider one or more of the following: obstacle uncertainty normalization index, importance of narrow passages or irreplaceable edges, edge prediction safety margin, and adjustment parameters for allocation scheduling. It can also further introduce a lower limit and upper limit for single collision event budgets to prevent a certain local event from excessively consuming the risk budget. Optionally, in this embodiment, when the actual risk of some edges is lower than the allocated budget, the unused risk budget can be placed in the reserve pool; when the edge of the narrow passage cannot pass under the original local allowable collision probability upper limit but still has sufficient geometric margin, the risk budget is transferred from the reserve pool and redistributed. If the redistributed risk budget exceeds the global allowable collision probability upper limit, the edge is rejected or an alternative path search is triggered.

[0043] The collision constraint module is used to set the collision probability between the swept body and the dynamically occupied obstacle body as the upper limit of the local allowed collision probability assigned to the edge of the corresponding candidate path, which is no greater than the collision probability limit.

[0044] In this embodiment, the local allowable collision probability upper limit can be calculated using methods such as distance-covariance sufficient condition, sampling upper confidence bound, interval envelope, reachability set envelope, and cumulative probability approximation of occupied grid. It is sufficient to be able to make a probabilistic safety determination of the edges of the candidate path based on uncertain obstacles and risk budget.

[0045] In this embodiment, optionally, when the obstacle's dynamic occupant is a capsule with an uncertain boundary, if the deterministic boundary of the sweeping body is not the lower limit of the obstacle's dynamic occupant's uncertain boundary (the lower limit refers to the uncertain boundary that encloses the smallest area of ​​the obstacle), then it can be determined that the collision probability between the sweeping body and the obstacle's dynamic occupant is not greater than the upper limit of the locally allowed collision probability assigned to the edge of the corresponding candidate path. In other embodiments, if the obstacle's dynamic occupant is constructed using an occupancy grid, the probability of the grid set covered by the sweeping body can be accumulated, and the proportion of the grid covered by the sweeping body can be controlled to not exceed a set proportion as a determination that no collision occurs.

[0046] The global planning module is used to construct multiple paths to be optimized based on the edges of candidate paths that satisfy collision constraints, and then obtain safe paths from the paths to be optimized through comprehensive cost optimization.

[0047] The global planning module adds the edges of candidate paths that satisfy the collision constraints to the global path graph and uses a global planning algorithm to obtain the path to be optimized. The comprehensive cost optimization includes one of the following: weighted summation, Pareto selection, multi-objective optimization, and hierarchical optimization.

[0048] The global planning algorithm used by the global planning module can be A. , Dijkstra, PRM, RRT, RRT-Connect, RRT FMT Methods such as CHOMP, TrajOpt, sampling optimization hybrid planning, and multi-planner orchestration only require combining the global allowed collision probability upper limit and the local allowed collision probability when judging whether the edges of the candidate path are feasible.

[0049] Optionally in this embodiment, after obtaining the safe path, the safe path can be further parameterized and smoothed over time, and then the collision constraints can be re-verified.

[0050] Optionally, in this embodiment, when the overall cost optimization adopts a weighted summation method, a fixed-weighted summation can be performed based on a combination of various sub-parameters such as execution time, estimated energy consumption, safe path length, joint change amount, used risk budget, safe path smoothness or slackness.

[0051] The security certificate module is used to generate security certificates. Security certificates are used to structurally record data related to obstacle dynamics, candidate paths, swept volumes, risk budget allocation, collision constraints, and safe paths.

[0052] Evidence for generating security certificates can be achieved using opportunity constraints, control barrier functions, reachability set analysis, interval analysis, probabilistic model testing, sampling confidence bounds, scenario tree verification, or formal rule engines.

[0053] Security certificates can be created using methods such as JSON, Protocol Buffers, binary structures, database records, blockchain-style hash chains, or controller-specific formats. Inconsistencies between execution parameters and security certificate records can also be prevented by verifying electronic signatures or version numbers.

[0054] In other embodiments, the structured data fields recorded in the security certificate may further include information related to real-time corrections made by the local security filtering module, information related to triggering degradation events by the degradation execution module, security path number, certificate validity period, certificate verification hash value / signature, etc. The security certificate serves four purposes: first, to prove to the multi-joint robotic arm's control system that the security path meets the corresponding collision probability upper limit before execution; second, to provide collision constraints to the multi-joint robotic arm's control system during its movements; third, to determine whether to allow continuation or execute degradation events such as speed limiting, replanning, or rollback when a risk event occurs; and fourth, to provide traceable evidence for subsequent auditing and maintenance. When the update of obstacle information leads to an increase in spatial pose covariance, a spatial mean offset exceeding a preset threshold, a joint configuration deviating too much from the record in the security certificate, or the security certificate expires, the security certificate is marked as invalid. At this time, the multi-joint robotic arm's control system does not continue to execute according to the original security certificate and triggers a replanning or degradation strategy.

[0055] The local safety filtering module is used to make real-time corrections to the multi-joint robotic arm without altering the safe path, ensuring that the robotic arm meets collision constraints during movement. The local safety filtering module corrects these collision constraints in real time by converting them into control constraints for the multi-joint robotic arm.

[0056] In this embodiment, real-time correction can be performed using methods such as CBF, MPC, QP, and reachability sets.

[0057] In this embodiment, an optional real-time correction method is safety function filtering; specifically, the safety margin between the robotic arm occupier and the obstacle dynamic occupier is constructed as a quadratic programming problem based on the distance from the boundary of the robotic arm occupier to the boundary of the obstacle dynamic occupier, the standard deviation of the normal uncertainty of the distance, and the motion margin of the multi-joint robotic arm.

[0058] In this embodiment, another optional real-time correction method is model prediction safety control; specifically, the system uses the positive kinematics model of the multi-joint robotic arm and the model prediction distribution of obstacle dynamics in the short prediction time domain to solve a set of control command sequences that satisfy collision constraints.

[0059] Optionally, in this embodiment, when the local safety filtering module fails within multiple consecutive cycles or the required braking distance exceeds the margin recorded in the safety certificate, the control system of the multi-joint robotic arm is forced to connect to the degraded execution module, which prioritizes speed limiting and replans the path. If local replanning is feasible, the safety certificate is replaced and execution continues. If local replanning is not feasible, the arm retracts to the previous joint configuration along the safe path recorded in the safety certificate. If retraction is also not feasible, a safety stop or safety emergency stop strategy is triggered.

[0060] The degraded execution module is used to set the safety level when the multi-joint robotic arm executes the corresponding safety path, and to control the multi-joint robotic arm to work in the corresponding action mode according to the safety level.

[0061] In this embodiment, the optional safety levels of the degraded execution module are as follows: Level 1 is warning and pre-deceleration, applicable to situations where obstacles are approaching but there is still sufficient safety margin; Level 2 is speed limit and local replanning, applicable to situations where the local risk of the safe path increases but there is still an alternative path; Level 3 is hold, rollback, or pause, applicable to situations where the safety certificate has expired or local cost optimization is not feasible; and Level 4 is safe stop or emergency stop, applicable to situations where personnel enter the danger zone, control communication fails, sensors fail, or braking margin is insufficient.

[0062] In addition to safety stops and retraction, the operation modes of multi-joint robotic arms can also include yielding, waiting, human-machine collaboration speed limiting, workstation light / sound and light reminders, switching to backup gripping points, switching to backup paths, or requesting the host computer to reschedule.

[0063] To further illustrate the specific composition of the system in this application, an implementation of the system based on a six-joint robotic arm is provided below, assuming the six-joint robotic arm is in a collaborative loading / unloading station avoiding dynamic personnel obstacles. The numerical values ​​in the following implementation are only for illustrating the corresponding implementation and are not intended to limit the scope of protection.

[0064] In the collaborative loading and unloading station, the planning cycle of the six-joint robotic arm is set to 0.10s, and the control cycle to 0.01s. The perception module connects to an RGB-D camera and a 3D LiDAR to acquire the outline, category, and spatial pose changes of obstacles for a person, and converts the obstacle information into structured state data. The structured state data includes the obstacle's mean position, velocity, covariance, category confidence, timestamp, validity period, and the model version number of the obstacle's dynamic occupant in the robotic arm's base coordinate system.

[0065] In this embodiment, the state vector of dynamic people and obstacles output by the perception module in the k-th planning cycle can be represented as X_k=[p_x,p_y,p_z,v_x,v_y,v_z] T Where p_x, p_y, and p_z represent the position coordinates of the obstacle in the robot arm's base coordinate system, and v_x, v_y, and v_z represent the obstacle's velocity vector. The perception module can use extended Kalman filtering, unscented Kalman filtering, or particle filtering to predict and update the obstacle's state. For example, it can predict the speed of the obstacle (person / obstacle) in the current planning cycle as 0.60 m / s, the total delay of perception and control as Delta_t as 0.03 s, and the maximum eigenvalue of the obstacle's position covariance matrix as lambda_max as 0.0004 m. 2 .

[0066] The obstacle model module constructs dynamic obstacles into capsule-shaped dynamic occupants based on the structured state data output by the perception module. The capsule's base radius r_0 is set to 0.10m, the base safety distance d_safe to 0.03m, the covariance confidence coefficient k_sigma to 2.0, and the upper limit of calibration error e_cal to 0.008m. The uncertainty expansion radius r_obs of the dynamic obstacle occupant is calculated using the following formula: r_obs=r_0+d_safe+k_sigma×sqrt(lambda_max)+|v|×Delta_t+e_cal; Substituting the above parameters, r_obs = 0.10 + 0.03 + 2.0 × sqrt(0.0004) + 0.60 × 0.03 + 0.008 = 0.196m. Therefore, the obstacle model module outputs a dynamic obstacle-occupying volume with uncertain boundaries. The output includes the capsule's centerline, radius (0.196m), timestamp, validity period, covariance, prediction speed, and model version number.

[0067] The candidate path module generates candidate nodes in the joint space based on the current and target joint configurations of the multi-joint robotic arm, and eliminates candidate nodes that violate joint limits, self-collision, and static safety envelopes. Let the current joint configuration be q_start=[0,-0.65,0.95,0,0.80,0], and the target joint configuration be q_goal=[1.05,-0.45,0.55,0.10,1.10,0.30]. The candidate path module generates two candidate paths: the first candidate path is a central narrow channel candidate path, formed by connecting q_start, q_N1, q_N2, and q_goal sequentially; the second candidate path is an outer bypass candidate path, formed by connecting q_start, q_O1, q_O2, and q_goal sequentially. The line connecting any two adjacent candidate nodes serves as an edge of a candidate path.

[0068] The robotic arm model module constructs the links of a six-joint robotic arm into robotic arm occupants with defined boundaries. Each link is approximated using a capsule, joints using a sphere, and the end effector using a bounding box. The equivalent radius r_arm of the robotic arm links is set to 0.06m. Based on the DH parameters, link dimensions, joint offsets, and candidate paths of the multi-joint robotic arm, the module performs joint interpolation on each edge of the candidate path and generates a robotic arm occupant under each interpolation configuration. Then, the robotic arm occupants on the same edge are aggregated to form the sweep body corresponding to that edge.

[0069] In calculating the collision probability, the collision threshold rho is taken as the sum of the uncertainty expansion radius of the dynamically occupied obstacle and the equivalent radius of the robotic arm link, i.e., rho = r_obs + r_arm = 0.196 + 0.06 = 0.256m. This collision threshold is used by the subsequent collision constraint module to determine whether the probabilistic safety requirements are met between the swept body and the dynamically occupied obstacle.

[0070] The risk budget module sets the global allowable collision probability cap between the swept body and the dynamically occupied obstacle as the risk budget. The global allowable collision probability cap epsilon_global is set to 0.03. For a candidate path consisting of edges of three candidate paths, the risk budget module uses a dynamic allocation mechanism to distribute the risk budget to the edges of the candidate path. The edge-level risk budget can be expressed as epsilon_i = epsilon_global × w_i, where epsilon_i is the upper limit of the local allowed collision probability corresponding to the edge of the i-th candidate path, w_i is the risk budget weight of the edge, and the sum of the weights of each edge is not greater than 1. For example, if the weights corresponding to the edges of the three candidate paths are w_1 = 0.25, w_2 = 0.50, and w_3 = 0.25, then the upper limits of the local allowed collision probability corresponding to the three edges are epsilon_1 = 0.03 × 0.25 = 0.0075, epsilon_2 = 0.03 × 0.50 = 0.015, and epsilon_3 = 0.03 × 0.25 = 0.0075, respectively.

[0071] The collision constraint module is used to determine whether the collision probability between the swept body and the dynamically occupied obstacle is not greater than the upper limit of the local allowed collision probability assigned to the edge of the corresponding candidate path. For the edge of the i-th candidate path, let the normal center distance between the swept body and the dynamically occupied obstacle be a random variable D_i, where D_i approximately follows a Gaussian distribution with mean mu_i and standard deviation sigma_i. Then the collision probability P_i of this edge can be expressed as P_i=P(D_i≤rho)=Phi((rho-mu_i) / sigma_i), where Phi is the standard normal distribution function, mu_i is the mean of the minimum center distance between the swept body and the dynamically occupied obstacle on this edge, and sigma_i is the corresponding standard deviation of uncertainty.

[0072] For the candidate path in the central narrow passage, on the edge of the second candidate path, the mean minimum center distance mu_2 between the robotic arm sweep body and the obstacle dynamic occupant is taken as 0.2718m, and the standard deviation sigma_2 is taken as 0.035m. Then P_2=Phi((0.256-0.2718) / 0.035)=Phi(-0.451), and the calculated P_2 is approximately 0.326. Since 0.326 is greater than the upper limit of the local allowable collision probability of 0.015 assigned to this edge, the second edge of the candidate path in the central narrow passage does not meet the collision constraint. The collision constraint module marks this edge as an infeasible edge, and the candidate path in the central narrow passage will not be included in the subsequent set of paths to be optimized.

[0073] For the candidate outer path, the collision probabilities corresponding to the edges of the three candidate paths are P_1, which is approximately 3.49 × 10. -10 P2 is approximately 3.47 × 102 -3 P_3 is approximately 4.57 × 10 -13 The three edges mentioned above satisfy P_1≤0.0075, P_2≤0.015, and P_3≤0.0075 respectively. Therefore, the edges of the three candidate paths of the outer bypass candidate path all satisfy the collision constraint, and the collision constraint module adds these three edges to the feasible edge set.

[0074] The global planning module constructs a global path graph based on the edges of candidate paths that satisfy collision constraints, and uses a global planning algorithm to obtain the path to be optimized. In this specific application example, the global planning module uses A... The path to be optimized is searched on the global path graph, and a weighted summation method is used for comprehensive cost optimization; the comprehensive cost function J can be expressed as follows: J=alpha×T+beta×L+gamma×R+eta×S; Where T is the path execution time, L is the path length or joint variation, R is the proportion of risk budget used, S is the path smoothness or jerk cost, and alpha, beta, gamma, and eta are weighting coefficients, for example, alpha=0.5, beta=0.3, gamma=0.1, and eta=0.1. Since the central narrow channel candidate path has edges that do not meet the collision constraints, while all edges of the outer bypass candidate path meet the collision constraints, and the comprehensive cost corresponding to the outer bypass candidate path meets the execution requirements, the global planning module outputs the outer bypass candidate path as a safe path.

[0075] After the global planning module outputs a safe path, the safety certificate module generates a safety certificate. The safety certificate structurally records the corresponding data for the obstacle dynamic occupant, candidate paths, swept volumes, risk budget allocation, collision constraints, and safe paths. In this specific application example, the data fields of the safety certificate include the safe path number, the obstacle dynamic occupant model version number, the obstacle dynamic occupant radius (0.196m), the equivalent radius of the robotic arm link (0.06m), the collision threshold (0.256m), the global allowed collision probability upper limit (0.03), the local allowed collision probability upper limit corresponding to the edges of each candidate path, the collision probability corresponding to the edges of each candidate path, the candidate node sequence of the safe path, the safety certificate validity period, and the certificate verification hash value or signature.

[0076] The local safety filtering module performs real-time corrections to the multi-joint robotic arm without altering the safe path, ensuring that the arm satisfies collision constraints during movement. This module transforms collision constraints into control constraints for the multi-joint robotic arm; the safety function is defined as h(q,t) = d(q,t) - rho, where d(q,t) is the real-time minimum center distance between the robotic arm's occupier and the obstacle's dynamic occupier under the current joint configuration q, and rho is the collision threshold. When h(q,t) ≥ 0, it indicates that the current state has a safety margin.

[0077] The local security filtering module can generate correction control commands using a quadratic programming approach: min 1 / 2×||u-u_ref|| 2 +M×delta 2 ; The constraints include h_dot(q,t,u)+a×h(q,t)≥-delta, u_min≤u≤u_max, and delta≥0; Where u_ref is the reference control command corresponding to the safe path, u is the corrected control command, delta is the relaxation variable, M is the relaxation variable penalty weight, and a is the safety convergence coefficient. In this way, the local safety filtering module makes real-time corrections to velocity, acceleration, or joint control quantities while minimizing deviation from the safe path.

[0078] For example, if after real-time perception updates it finds that d(q,t) = 0.270m and the collision threshold rho = 0.256m at a certain moment, then h(q,t) = 0.014m. The local safety filtering module determines that there is still a safety margin, but the safety margin has decreased. Therefore, it performs speed limit correction on the reference control command to make the multi-joint robotic arm decelerate along the safe path. When real-time perception updates cause h(q,t) < 0, or when the secondary planning is infeasible for multiple consecutive control cycles, the local safety filtering module outputs an infeasibility flag.

[0079] The degraded execution module sets the safety level based on the safety certificate status, the results of the local safety filtering module, and the real-time risk assessment results, and controls the multi-joint robotic arm to work in the corresponding action mode. When the real-time collision probability is close to but does not exceed the local allowable collision probability limit, the first-level safety level is triggered, and prompts and pre-deceleration are executed; when the local risk increases but there is an alternative path, the second-level safety level is triggered, and speed limits and local replanning are executed; when the safety certificate expires, the local safety filtering module is infeasible for three consecutive control cycles, or local cost optimization is infeasible, the third-level safety level is triggered, and hold, retreat, or pause are executed; when personnel enter the danger zone, control communication fails, sensors fail, or braking margin is insufficient, the fourth-level safety level is triggered, and a safe stop or emergency stop is executed.

[0080] Through the above specific implementation method description, the multi-joint robotic arm path planning system in this embodiment can realize the complete operation logic of "the perception module acquires obstacle information, the obstacle model module generates dynamic obstacle occupants, the candidate path module generates candidate paths, the robotic arm model module generates swept bodies, the risk budget module allocates the upper limit of local allowable collision probability, the collision constraint module filters feasible edges, the global planning module outputs safe paths, the safety certificate module records auditable data, the local safety filtering module performs real-time correction, and the degraded execution module performs safety level control". The output path planning result includes the safe path, specific algorithm, specific parameters, collision probability, upper limit of local allowable collision probability, risk budget allocation result, and safety correction logic in the execution stage.

[0081] Compared with the prior art, the advantages of this application are as follows: Instead of relying solely on deterministic geometric distances to determine path safety, this approach provides quantifiable safety credentials for each path and its edges. It reduces unnecessary boundary occupancy and improves traffic efficiency when the uncertainty in measuring dynamic obstacles is low; and expands boundary occupancy to avoid insufficient safety margins when the uncertainty in measuring dynamic obstacles is high or dynamic risks increase. This enhances the safety, interpretability, and engineering deployability of planned paths in human-machine collaboration, open production lines, dynamic sorting, and high-value automated equipment. The obstacle model is expanded from a fixed geometry to an occupier with an uncertain boundary, including mean, covariance, confidence level, and validity period. Measurement noise, calibration error, occlusion ratio, tracking residual, and communication latency are converted into relevant parameters of the uncertain boundary that can be used for path planning, adapting to actual production scenarios with a variety of dynamic objects as obstacles. For risk budget allocation of the entire path, collision constraints are replaced with simple fixed distance thresholds, so that the safety of the planned path can be introduced by the collision probability representation. Combined with the method of judging the collision probability between the sweep body and the dynamic occupier of the obstacle on the edge of the candidate path, a provable safety basis is provided for global path planning. A safety certificate is provided to offer a reliable data anchor for the control of the multi-joint robotic arm, enhancing the data security of path planning. Local safety filtering is then used to further control the local movements of the multi-joint robotic arm, preventing path failure or planning failure due to tracking errors, sensor updates, or changes in dynamic obstacles. A tiered degradation execution method is employed, where the system automatically limits speed, replans locally, reverts along the safety certificate, or triggers a safety stop when opportunity constraints cannot be met or local safety optimization is not feasible, ensuring the hardware safety of the multi-joint robotic arm.

[0082] like Figure 2 As shown in the figure, this application discloses a path planning method for a multi-joint robotic arm based on uncertainty perception, which includes the following steps.

[0083] S1. Obtain obstacle information corresponding to the multi-joint robotic arm; S2. Based on obstacle information, construct obstacles into dynamic obstacle occupants with uncertain boundaries; S3. Generate candidate nodes in the joint space of the multi-joint robotic arm, and then remove candidate nodes that violate joint limits, self-collision, and static safety envelope to obtain candidate paths for the multi-joint robotic arm. S4. After constructing the multi-joint robotic arm as a robotic arm occupying body with defined boundaries, generate the sweeping body of the multi-joint robotic arm based on the robotic arm occupying body and candidate paths. S5. Set the global allowed collision probability upper limit between the swept body and the dynamically occupied obstacle body as the risk budget, and then allocate the risk budget to the edges of the candidate path as the corresponding local allowed collision probability upper limit. S6. Set the collision probability between the swept body and the dynamically occupied obstacle body as the upper limit of the local allowed collision probability assigned to the edge of the corresponding candidate path as a collision constraint. S7. Construct multiple paths to be optimized based on the edges of candidate paths that satisfy collision constraints, and then obtain safe paths from the paths to be optimized through comprehensive cost optimization. S8. Structured recording of obstacle dynamic occupancy, candidate paths, swept volumes, risk budget allocation, collision constraints, and safe paths; S9. Without changing the safe path, make real-time corrections to the multi-joint robotic arm so that the multi-joint robotic arm meets collision constraints during operation. S10. Set the safety level when the multi-joint robotic arm executes the corresponding safety path, and control the multi-joint robotic arm to work in the corresponding action mode according to the safety level.

[0084] The methods described above are executed based on the corresponding uncertainty-aware multi-joint robotic arm path planning systems in the foregoing embodiments, and have the beneficial effects of the corresponding system embodiments, which will not be repeated here.

[0085] like Figure 3 As shown, based on the same inventive concept, corresponding to any of the above embodiments, this application also discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the above-described multi-joint robotic arm path planning method based on uncertainty perception.

[0086] Specifically, the device includes: a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, memory 1020, input / output interface 1030, and communication interface 1040 are interconnected within the device via the bus 1050.

[0087] The processor 1010 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), GPU (Graphics Processing Unit), or one or more integrated circuits, to implement relevant programs and achieve the technical solutions provided in the embodiments of this specification.

[0088] The memory 1020 can be implemented in the form of ROM (Read-Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 1020 and called by the processor 1010. The input / output interface 1030 is used to connect input / output modules to realize information input and output. Input / output modules can be configured as components in the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touch screens, microphones, various sensors, etc., and output devices may include displays, projectors, speakers, vibrators, indicator lights, etc.

[0089] The communication interface 1040 is used to connect the communication module (not shown in the figure) to enable communication between this device and other devices. The communication module can communicate via wired means (such as USB (Universal Serial Bus), network cable, etc.) or wireless means (such as mobile network, WIFI (Wireless Fidelity), Bluetooth, etc.).

[0090] Bus 1050 includes a pathway for transmitting information between various components of the device, such as processor 1010, memory 1020, input / output interface 1030, and communication interface 1040.

[0091] It should be noted that although the above-described device only shows the processor 1010, memory 1020, input / output interface 1030, communication interface 1040, and bus 1050, in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the embodiments of this specification, and not necessarily all the components shown in the figures.

[0092] The electronic device described above is used to implement the corresponding uncertainty-aware multi-joint robotic arm path planning method in the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0093] Based on the same inventive concept, corresponding to any of the above embodiments, this application also discloses a computer-readable storage medium that stores computer instructions for enabling a computer to implement the above-described multi-joint robotic arm path planning method based on uncertainty perception.

[0094] The computer-readable storage medium of this embodiment includes permanent and non-permanent, removable and non-removable media that can be implemented by any system or technology for information storage. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device. The computer instructions stored in the storage medium of the above embodiments are used to enable the computer to implement the uncertainty-aware multi-joint robotic arm path planning method as described in the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0095] The above description is merely a preferred embodiment and the technical principles employed in this application. This application is not limited to the specific embodiments or combinations thereof, and various obvious changes, readjustments, and substitutions that can be made by those skilled in the art will not depart from the scope of protection of this application. Therefore, although this application has been described in detail through the above embodiments, this application is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of this application, the scope of which is determined by the scope of the claims.

Claims

1. A path planning system for a multi-joint robotic arm based on uncertainty perception, characterized in that, include: The perception module is used to acquire obstacle information for the multi-joint robotic arm; An obstacle model module is used to construct obstacles as dynamic obstacle occupants with uncertain boundaries based on the obstacle information. The candidate path module is used to generate candidate nodes in the joint space of the multi-joint robot arm, and then remove candidate nodes that violate joint limits, self-collision, and static safety envelope to obtain candidate paths for the multi-joint robot arm. The robotic arm model module is used to construct a multi-joint robotic arm as a robotic arm occupant with defined boundaries, and then generate a sweep body of the multi-joint robotic arm based on the robotic arm occupant and the candidate path. The risk budget module is used to set the global allowable collision probability upper limit between the swept body and the obstacle dynamic occupier as the risk budget, and then allocate the risk budget to the edges of the candidate path as the corresponding local allowable collision probability upper limit. The collision constraint module is used to set the collision probability between the swept body and the obstacle dynamic occupier as the upper limit of the local allowed collision probability to which the edge of the corresponding candidate path is assigned, which is not greater than the collision probability limit of the obstacle dynamic occupier. The global planning module is used to construct multiple paths to be optimized based on the edges of candidate paths that satisfy the collision constraints, and then obtain a safe path from the paths to be optimized through comprehensive cost optimization.

2. The multi-joint robotic arm path planning system based on uncertainty perception according to claim 1, characterized in that, It also includes a security certificate module for generating security certificates; The safety certificate is used to structurally record data such as obstacle dynamic occupants, candidate paths, swept volumes, risk budget allocation, collision constraints, and the corresponding data of the safety paths. And / or, it also includes a local safety filtering module, which is used to make real-time corrections to the multi-joint robotic arm without changing the safety path, so that the multi-joint robotic arm meets collision constraints when it moves. And / or, it also includes a degraded execution module, used to set the safety level when the multi-joint robotic arm executes the corresponding safety path, and control the multi-joint robotic arm to work in the corresponding action mode according to the safety level.

3. The multi-joint robotic arm path planning system based on uncertainty perception according to claim 1, characterized in that, The local safety filtering module corrects collision constraints in real time by converting them into control constraints for the multi-joint robotic arm.

4. The multi-joint robotic arm path planning system based on uncertainty perception according to claim 1, characterized in that, The obstacle dynamic occupancy volume is constructed using one or more of the following methods: ellipsoid, sphere, capsule, bounding box, occupancy grid, signed distance field, Gaussian object, occupancy octree, interval envelope, fuzzy set, reachability set, and neural network prediction.

5. The multi-joint robotic arm path planning system based on uncertainty perception according to claim 1, characterized in that, The robotic arm's occupying body can be one or more of the following: capsule, sphere, bounding box, and mesh. The swept body is formed by the robotic arm occupying the body through the candidate path.

6. The multi-joint robotic arm path planning system based on uncertainty perception according to claim 1, characterized in that, The allocation of risk budget adopts a dynamic allocation mechanism.

7. The multi-joint robotic arm path planning system based on uncertainty perception according to claim 1, characterized in that, Edges of candidate paths that satisfy collision constraints are added to the global path graph, and a global planning algorithm is used to obtain the path to be optimized. The comprehensive cost optimization includes one of the following: weighted summation, Pareto selection, multi-objective optimization, and hierarchical optimization.

8. A path planning method for a multi-joint robotic arm based on uncertainty perception, characterized in that, include: Obtain obstacle information corresponding to the multi-joint robotic arm; Based on the obstacle information, the obstacles are constructed as dynamic obstacle occupants with uncertain boundaries; Candidate nodes are generated in the joint space of the multi-joint robotic arm, and then candidate nodes that violate joint limits, self-collision, and static safety envelope are eliminated to obtain candidate paths for the multi-joint robotic arm. After constructing the multi-joint robotic arm as a robotic arm occupant with defined boundaries, the sweep body of the multi-joint robotic arm is generated based on the robotic arm occupant and the candidate path. Set the global allowed collision probability upper limit between the swept body and the obstacle dynamic occupier as the risk budget, and then allocate the risk budget to the edges of the candidate path as the corresponding local allowed collision probability upper limit. The collision constraint is set as the upper limit of the local allowed collision probability, which is no greater than the collision probability of the edge assigned to the corresponding candidate path. Multiple paths to be optimized are constructed based on the edges of candidate paths that satisfy the collision constraints, and then a safe path is obtained from the paths to be optimized through comprehensive cost optimization.

9. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the uncertainty-aware multi-joint robotic arm path planning method of claim 8.

10. A storage medium, which is a computer-readable storage medium, characterized in that, The computer contains computer instructions that enable the computer to implement the uncertainty-aware multi-joint robotic arm path planning method of claim 8.

Images