A method for online adaptation and execution of robot trajectories oriented to action blocks

Abstract

The application discloses a kind of action block-oriented robot trajectory online adaptation and execution method, belong to robot motion control technical field, comprising: after separating mechanical arm dimension and gripper dimension from action block, respectively with current robot state splicing, form original mechanical arm trajectory and original gripper trajectory;Path parameters are established based on the cumulative execution distance between adjacent path points in original mechanical arm trajectory, and original mechanical arm trajectory is time parameterized in combination with physical constraints, to obtain the mechanical arm trajectory after retiming;According to the path parameters corresponding to the mechanical arm trajectory after retiming, original gripper trajectory is synchronously interpolated, to obtain the gripper trajectory after retiming, and the mechanical arm trajectory and gripper trajectory after retiming are sampled according to robot control cycle, to generate retiming execution trajectory.The application can significantly improve the efficiency of robot action execution, optimize motion smoothness and ensure multi-dimensional action collaboration, improve the stability and reliability of online deployment.

Description

Technical Field

[0001] This invention belongs to the field of robot motion control technology, specifically relating to a method for online adaptation and execution of robot trajectories oriented towards action blocks. Background Technology

[0002] Vision-Language-Action Model (VLA) is an important direction for achieving advanced robot control. It can directly output robot actions based on visual observations, natural language commands, and the robot's current state. To make the actions smoother and enable short-time planning, mainstream VLA models often adopt an action block strategy, that is, outputting a sequence of robot actions for multiple future time steps at once. However, the action blocks output by these models are essentially discrete, equally timed sequences of commands, without explicitly considering the physical constraints of the robot itself. Therefore, in actual execution, they often fail to fully utilize the robot's dynamic performance.

[0003] In the field of robot motion control, techniques for time parameterization of known geometric paths (such as a series of path points in joint space) already exist. For example, the Time-Optimal Path Parameterization (TOPP) algorithm can calculate a time-optimal or executable time-varying trajectory that satisfies the constraints given physical constraints such as robot joint velocity and acceleration. However, in existing typical VLA schemes, a naive point-by-point action block playback method is still used for deployment on simulation or real robots. Its execution flow is as follows: the model receives the current image and instructions, and outputs an action block containing K future actions; the robot's underlying control system executes each action in the action block sequentially according to a fixed control cycle; after executing the entire action block, the model acquires observations again and outputs the next action block, and so on.

[0004] The execution process of this scheme can be understood as treating the action block as a set of discrete command sequences, and performing point-by-point playback in an open-loop or short-time open-loop manner. It has the following obvious drawbacks: (1) Execution efficiency is limited by the distribution of teaching data: VLA models are usually trained based on human teleoperation data. The rhythm of human teaching is conservative and slow, which causes the action blocks learned by the model to inherit this slow characteristic and fail to make full use of the high-speed and high-dynamic motion capabilities actually possessed by the robot.

[0005] (2) Inability to utilize the entire path information for optimization: Action blocks are executed in isolation, point by point, with each path point serving only as an independent control instruction. This makes it impossible to utilize the geometric path structure formed by the entire action block for time allocation and motion optimization from a global perspective.

[0006] (3) Timing mismatch between motion dimensions: When the motion trajectory of the robotic arm is time-compressed or optimized, if the action of the gripper is still executed according to the discrete time index of the original action block, there will be timing disorder problems such as the gripper not closing in place before the robotic arm starts to move, or the robotic arm not closing in place before the gripper opens in advance.

[0007] Therefore, there is an urgent need for a robot motion control method that can perform global time optimization of action blocks and ensure multi-dimensional action timing coordination. Summary of the Invention

[0008] In view of the above, the purpose of this invention is to provide an online adaptation and execution method for robot trajectories oriented towards action blocks. The action block is regarded as a geometric path to be optimized, and it is retiming through online time parameterization to generate a retiming execution trajectory that satisfies the robot's physical constraints and has a faster rhythm. This significantly improves the efficiency of motion execution, optimizes motion smoothness, and ensures the synchronization of gripper action and the new rhythm of the robotic arm while satisfying the robot's physical constraints. Furthermore, a built-in failure rollback mechanism is incorporated to realize a complete and closed-loop online execution process, thereby improving the coordination and robustness of online motion control and deployment of embodied intelligent robots.

[0009] To achieve the above-mentioned objectives, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for online adaptation and execution of robot trajectories oriented towards action blocks, comprising the following steps: Obtain an action block containing multiple future path points. Separate the robotic arm dimension and gripper dimension from the action block and then concatenate them with the current robot state to form the original robotic arm trajectory and the original gripper trajectory. Path parameters are established based on the cumulative execution distance between adjacent path points in the original robotic arm trajectory. The original robotic arm trajectory is then time-parameterized based on the path parameters and physical constraints to obtain the retimed robotic arm trajectory. The original gripper trajectory is synchronously interpolated based on the path parameters corresponding to the retiming robotic arm trajectory to obtain the retiming gripper trajectory. The retiming robotic arm trajectory and gripper trajectory are sampled according to the robot control cycle to generate the retiming execution trajectory to control the robot execution.

[0010] Preferably, the step of establishing path parameters based on the cumulative execution distance between adjacent path points in the original robotic arm trajectory includes: Calculate the cumulative joint space weighted distance between adjacent path points in the original robotic arm trajectory, or the cumulative end effector movement distance in Cartesian space, or a weighted combination of both, to obtain the monotonically increasing path parameter value corresponding to each path point. This path parameter value represents the cumulative arc length from the trajectory starting point along the geometric path to the current path point.

[0011] Preferably, before establishing path parameters, the original robotic arm trajectory is further deduplicated: the starting point and ending point of the original robotic arm trajectory are retained, and the remaining path points are traversed to remove duplicate points that are the same as the most recent retained path point, and near duplicate points whose weighted distance from the most recent retained path point is less than a preset threshold are removed.

[0012] Preferably, the step of time parameterizing the original robotic arm trajectory based on path parameters and physical constraints to obtain the retimed robotic arm trajectory includes: The path parameter increment between adjacent grid points is determined at each grid point. The velocity square of the discrete path parameter to be solved is defined at each grid point, and the acceleration of the discrete path parameter to be solved is defined between adjacent grid points. Based on the path parameter increment, the velocity square of the discrete path parameter, and the acceleration of the discrete path parameter, the discrete dynamic equation of the velocity square between adjacent grid points is established. The physical constraints on the joint velocity and acceleration of the robotic arm are transformed into inequality constraints on the discrete path parameter velocity square and discrete path parameter acceleration. The discrete dynamic equations are solved under the inequality constraints to obtain the discrete path parameter velocity square value sequence and the corresponding acceleration value sequence. The time increment corresponding to the incremental path parameter for each path segment is calculated based on the discrete path parameter velocity square value sequence and the discrete path parameter acceleration sequence. The time increments are accumulated sequentially to obtain the retiming time at each grid point. The execution time of each path point of the original robotic arm trajectory is updated according to the retiming time to form the retiming robotic arm trajectory.

[0013] Preferably, the step of synchronously interpolating the original gripper trajectory based on the path parameters corresponding to the retiming robotic arm trajectory to obtain the retiming gripper trajectory includes: Establish the corresponding robotic arm path parameter position for each state point in the original gripper trajectory, and construct an interpolation function of the gripper state with respect to the robotic arm path parameters based on the robotic arm path parameter position; Based on the path parameter values ​​of the robotic arm trajectory at any time after retiming, the gripper state at that time is calculated through an interpolation function, so that the gripper movement is synchronized with the progress of the robotic arm along the path.

[0014] Preferably, the retiming execution trajectory includes: a sequence of robot arm joint positions, a sequence of robot arm joint speeds, and a sequence of gripper states.

[0015] Preferably, it also includes a failure rollback step: when time parameterization fails, it automatically rolls back to the point-by-point execution mode of the original action block and executes each original path point in the action block in sequence.

[0016] Secondly, the embodiments of the present invention provide a robot trajectory online adaptation and execution system oriented towards action blocks, which is implemented using the above-mentioned robot trajectory online adaptation and execution method oriented towards action blocks, including: action block acquisition module, path construction module, path parameter establishment module, trajectory time parameterization module, gripper synchronization module, control sequence generation module, and backtracking execution module; The action block acquisition module is used to acquire action blocks that contain multiple future path points; The path construction module is used to separate the robotic arm dimension and gripper dimension from the action block and then splice them with the current robot state to form the original robotic arm trajectory and the original gripper trajectory. The path parameter establishment module is used to establish path parameters based on the cumulative execution distance between adjacent path points in the original robotic arm trajectory. The trajectory time parameterization module is used to perform time parameterization on the original robotic arm trajectory based on path parameters and physical constraints to obtain the robotic arm trajectory after retiming. The gripper synchronization module is used to synchronously interpolate the original gripper trajectory based on the path parameters corresponding to the retiming robotic arm trajectory, so as to obtain the retiming gripper trajectory. The control sequence generation module is used to sample the retiming trajectory of the robotic arm and gripper according to the robot control cycle, and generate the retiming execution trajectory to control the robot to execute. The rollback execution module is used to automatically roll back to the point-by-point execution mode of the original action block when time parameterization fails, and execute each original path point in the action block in sequence.

[0017] Thirdly, an electronic device provided by an embodiment of the present invention includes a memory and one or more processors. The memory is used to store a computer program, and the processor is used to implement the above-described method for online adaptation and execution of robot trajectories oriented towards action blocks when the computer program is executed.

[0018] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program, which, when executed by a computer, implements the above-described method for online adaptation and execution of robot trajectories oriented towards action blocks.

[0019] Compared with the prior art, the beneficial effects of the present invention include at least the following: (1) Improve execution efficiency: This invention treats conservative action blocks as geometric paths to be optimized and performs retiming through online time parameterization. Under the premise of satisfying the physical constraints of robot joint velocity and acceleration, the original action sequence is time-compressed, which significantly shortens the execution time (specifically, the total trajectory time after retiming). T(significantly shorter than the conservative execution time of the original action block), thereby breaking through the inherent conservative rhythm limitations in human teaching data and fully releasing the high dynamic motion capabilities of the robot body.

[0020] (2) Optimize motion smoothness: Based on the geometric structure of the entire path, the present invention performs unified time parameterization, which can generate a continuous retiming execution trajectory of velocity and acceleration, avoiding the motion impact that may be caused by point-by-point playback execution, making the robot motion smoother and more stable.

[0021] (3) Ensure multi-dimensional motion coordination: This invention maps the gripper state to the path parameter position for synchronous interpolation, so that the gripper opening and closing action is strictly matched with the progress of the robotic arm path. Even if the execution rhythm of the robotic arm is accelerated or adjusted, the timing disorder can be eliminated, ensuring the accuracy and consistency of the operation.

[0022] (4) Improve system robustness: The present invention has a built-in failure rollback mechanism. When the time parameterization solution fails (such as path degradation, no feasible solution, etc.), it automatically switches to the point-by-point execution mode of the original action block, which avoids task interruption or loss of control and enhances the stability and reliability in the online deployment environment. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present 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 only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart illustrating the online adaptation and execution method for robot trajectories oriented towards action blocks provided in an embodiment of the present invention. Figure 2 This is a comparison chart of the results of the present invention and the baseline method in terms of path advancement rate; Figure 3 This is a comparison chart of the results of the present invention and the baseline method in terms of velocity envelope; Figure 4 This is a comparison chart of the results of the present invention and the baseline method in terms of acceleration envelope; Figure 5 This is a schematic diagram of the structure of the robot trajectory online adaptation and execution system oriented towards action blocks provided in an embodiment of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not limit the scope of protection of this invention.

[0026] The inventive concept of this invention is as follows: Addressing the problems in existing technologies where the execution efficiency of action blocks is limited by teaching data, the inability to utilize the entire path for global time allocation, and the tendency for timing mismatch between multi-dimensional actions (such as the robotic arm and gripper), this invention provides an online adaptation and execution method for robot trajectories oriented towards action blocks. This method involves concatenating action blocks with the current robot state to form an original geometric path, establishing path parameters based on the cumulative execution distance between adjacent path points, and retiming the robotic arm path online using time parameterization while satisfying the physical constraints of robot joint velocity and acceleration. Furthermore, it involves synchronously interpolating the gripper trajectory based on the retimed path parameters. This significantly improves action execution efficiency, optimizes motion smoothness, and ensures the timing coordination of the robotic arm and gripper actions.

[0027] like Figure 1 As shown, the embodiment provides a method for online adaptation and execution of robot trajectories oriented towards action blocks, including the following steps: Step S101: Obtain an action block containing multiple future path points.

[0028] In this embodiment, robot observation information, task language instructions, and the current robot state (including the current state of the robotic arm joints) are acquired. and current gripper status The input is a Visual Language Action Model (VLA), and the model output includes... Action block for a future path point In other embodiments, action blocks The acquisition is not limited to the VLA model, but also includes the use of diffusion strategy models, traditional trajectory planners, or any module that can output future discrete action sequences.

[0029] Step S102: After separating the robotic arm dimension and gripper dimension from the action block, they are spliced ​​with the current robot state to form the original robotic arm trajectory and the original gripper trajectory.

[0030] In the embodiment, from the action block Each path point in Separate the dimensions of the robotic arm and gripper dimensions : , in, This represents the robot joint angles or end-effector pose components at path points, used to construct the geometric path. This indicates the gripper opening / closing degree or force control command component at the path point. This represents the path point index in the original action block. Represents the space of all real numbers. Indicates the degrees of freedom of the robotic arm. This indicates the degrees of freedom of the gripper.

[0031] For dual-arm robots It can be further separated into the left robotic arm dimension. And the right robotic arm dimension Then, subsequent time parameterization and gripper synchronous interpolation are performed on the left and right dimensions respectively: .

[0032] Obtain the current joint state of the robot's robotic arm and current gripper status The current state is then concatenated with the dimensions of the action block to form a complete original robotic arm trajectory (i.e., a sequence of robotic arm path points). And the original gripper trajectory (i.e., the gripper state sequence). : , , in, Represents the first in the original robotic arm trajectory Path points, Represents the first in the original gripper trajectory There are path points.

[0033] Step S103: Establish path parameters based on the cumulative execution distance between adjacent path points in the original robotic arm trajectory.

[0034] In the embodiments, for Perform a traversal, removing duplicates or path points with excessively close weighted Euclidean distances. Define the weighted norm. : , in, Represents the following formula , This represents a diagonal weight matrix, where each diagonal element is used to normalize the dimensions of different joints (i.e., adjust the contribution weights of different joint distances). The superscript... This indicates transpose. The pathpoint preservation rules are as follows: , , in, This indicates the most recently saved pathpoint. This represents the preset near-repetition threshold. The retained sequence of robotic arm pathpoints. Recorded as: , in, This indicates the number of characters retained after deduplication. Each robotic arm path point This indicates the number of robotic arm path points retained after deduplication.

[0035] Record-preserving original index mapping ,satisfy .

[0036] Calculate the joint space distance between adjacent path points : , Establish monotonically increasing path parameters (arc length parameters): , in, Represents the minimum path increment (which is a very small constant, such as...) This is used to prevent path parameter degradation and ensure that the path parameters are strictly monotonically increasing. This yields the path parameter sequence. : , in, Indicates the first The path parameter value, that is, along the geometric path from the starting point to the nth path. The cumulative arc length of each point. This represents the total arc length of the entire geometric path (i.e., the upper bound of the path parameter range).

[0037] The geometric path of a robotic arm can be represented as path parameters. Functions: , in, This indicates that the path parameters will be used. A continuous function mapped to a joint space location can be achieved through piecewise linear interpolation or spline interpolation.

[0038] Step S104: Based on path parameters and physical constraints, perform time parameterization on the original robotic arm trajectory to obtain the retiming robotic arm trajectory.

[0039] In this embodiment, the robotic arm geometric path obtained in step S103 is used. and path parameter sequence Based on this, the path parameters are solved under robot kinematic constraints using a time-parameterized algorithm. With time Optimal mapping . This represents a time parameterization function, i.e., real time. Mapping to path parameters It is a monotonically non-decreasing function.

[0040] Define the first derivative of the path parameter and second derivative : , in, Indicates the path parameter speed. This represents the path parameter acceleration.

[0041] According to the chain rule, the joint speed of the robotic arm and acceleration It can be represented as: , , Wherein, the first derivative of the joint position with respect to the path parameters The second derivative of joint position with respect to path parameters .

[0042] For each joint The velocity and acceleration constraints are as follows: , in, This represents the upper limit of joint velocity, i.e., the first... The maximum permissible movement speed of each joint is derived from the robot's body parameters; This represents the upper limit of joint acceleration, i.e., the first... The maximum allowable acceleration of each joint is derived from the robot's body parameters; This indicates the total number of joints.

[0043] Substituting into the chain rule expression, we get: , , In the discrete form, at the path parameter grid points Definition above: , , in, This represents the square of the discrete path parameter velocity, i.e., at the path parameter grid points. The square of the velocity at that point; This represents the path parameter increment, i.e., the difference between adjacent path parameter points.

[0044] Let discrete path parameter acceleration For interval constant on If the value is given, then the discrete dynamic equation is: , We need to find the sequence that satisfies the following conditions and : , , , , in, This indicates the path parameter speed at the starting point of the path. This indicates the path parameter speed at the end of the path.

[0045] For typical stop-stop action block execution, the boundary velocity is usually set to zero: .

[0046] Obtain the optimal solution and Then, the time increment between adjacent path points and retiming time The calculation is as follows: , , in, This represents the time required for the robotic arm to move between two adjacent deduplication path points. Represents the deduplicated path points of the robotic arm. The execution time corresponding to the retiming trajectory .

[0047] The total duration of the trajectory after retiming is: , in, This represents the total execution time of the entire trajectory from the start point to the end point after retiming.

[0048] In other embodiments, solving and This method can be replaced by other time-parameterized algorithms, such as trajectory generation based on convex optimization, model predictive control (MPC), or trapezoidal velocity curve planning. Correspondingly, the discrete dynamic equations... It can also be replaced with other integral formats.

[0049] Step S105: Synchronously interpolate the original gripper trajectory based on the path parameters corresponding to the retiming robotic arm trajectory to obtain the retiming gripper trajectory.

[0050] This step is crucial to the invention. The gripper state is no longer based on the original discrete index. Instead of executing, it follows the path parameters after the three-step timing process. Synchronize.

[0051] First, the original gripper sequence Each point in Establish the corresponding path parameter location As anchor points. One implementation is to use the original robotic arm path points. Projected onto the deduplicated path superior: , , in, Represents the original robotic arm path point sequence Middle and original gripper sequence points The index of the nearest path point, Represents the nearest neighbor path point The corresponding path parameter value (i.e., cumulative arc length).

[0052] Another implementation method is based on the cumulative distance before deduplication: , , , in, This represents a truncation function (or limiting function), which restricts the input value to a specified minimum and maximum range, i.e., the calculated path parameters. Limited to the range of valid paths To prevent out-of-bounds errors, the minimum value should be output if the input value is less than the minimum value, and the maximum value should be output if the input value is greater than the maximum value; otherwise, the original value should be left unchanged.

[0053] Construct gripper state with respect to path parameters function For continuous grippers, linear interpolation is used: , For discrete grippers (such as open / closed states), piecewise interpolation is used: , For the retimed trajectory at any given time Path parameters The gripper state at that moment is: .

[0054] In other embodiments, for discrete grippers, nearest neighbor synchronization can also be used: Alternatively, synchronization can be achieved using an event-triggered method: when the path parameter... Exceeding the threshold The state switch is triggered at any time.

[0055] Step S106: Sample the retiming trajectory of the robotic arm and gripper according to the robot control cycle, and generate the retiming execution trajectory to control the robot to execute.

[0056] Set the robot's control cycle as (Set to 0.001s in this embodiment), the control frequency is Total number of sampling points (i.e., the number of control commands obtained after sampling the retime trajectory according to the control cycle). for: , The sampling time points are: , At each sampling time Calculate the corresponding path parameters : , Generate retiming execution trajectories (i.e., high-frequency execution sequences), including: Joint position sequence: , indicating at the sampling time The target joint position is sent to the robot; Joint velocity sequence: , indicating at the sampling time The joint target velocities sent to the robot, where From discrete solutions Interpolation is obtained; Gripper state sequence: , indicating at the sampling time The target status of the gripper is sent to the robot.

[0057] Final executable trajectory The complete trajectory data structure, containing all high-frequency instructions and metadata, is represented as follows: , in, This represents the sequence of robotic arm joint positions that are densely sampled according to the control cycle after retiming. This represents the sequence of robot arm joint velocities that are densely sampled according to the control cycle after retiming. This represents the gripper state sequence that is densely sampled according to the control cycle after retiming. This indicates the total execution time of the trajectory after retiming (or the trajectory time parameter after retiming). This represents the sequence of retimed execution times corresponding to the critical path points after deduplication.

[0058] Step S107: When time parameterization fails, automatically revert to the point-by-point execution mode of the original action block and execute each original path point in the action block in sequence.

[0059] In the embodiment, if an error occurs in any of the above steps, for example: , If the number of path points is insufficient, the path parameter increment is non-positive, there is no feasible solution that satisfies the physical constraints, or valid sampling points cannot be generated, then the time parameterization is deemed a failure. The system will automatically revert to the point-by-point execution mode of the original action block, that is, execute sequentially. This is to ensure that the robot's online operation is uninterrupted.

[0060] Furthermore, in the place_phone_stand task of the RoboTwin dataset, the baseline scheme (π0.5 type check model, source file name π0.5: a Vision-Language-Action Model with Open-World Generalization) and the method of this invention were verified using the same VLA model checkpoints, the same task configuration, and the same number of tests. The results are shown in Table 1.

[0061] Table 1 Verified Results

[0062] The comparison results in Table 1 show that: (1) Execution time: The total execution time of the baseline point-by-point execution is 35.59 seconds, while the method of the present invention (online retiming) only requires 4.35 seconds, which reduces the execution time by about 87.8%. This shows that the method of the present invention can make full use of the entire path information for global time optimization, significantly improve the execution efficiency under the premise of satisfying the robot's physical constraints, and break through the inherent conservative rhythm limitation in human teaching data.

[0063] (2) Number of trajectory samples: The baseline method generated 8899 control command points, while this method only generated 4348 sampling points, a reduction of about 51.1%. Fewer sampling points mean lower control communication burden and computational resource consumption. At the same time, since the trajectory generated by this method is smoother and faster, the actual execution effect is better.

[0064] (3) Average velocity RMS (root mean square of joint velocity): The average velocity RMS of the baseline method is only 0.056 rad / s, reflecting its slow movement and frequent pauses; the average velocity RMS of this method is increased to 0.468 rad / s, which is about 8.4 times that of the baseline method. This shows that this method can actively utilize the robot's high dynamic motion capability to enable the robotic arm to maintain a higher motion rhythm, thereby significantly shortening the task completion time.

[0065] (4) Retiming overhead: The baseline method requires no retiming calculation, with an overhead of 0.0 ms; the average overhead of online retiming introduced by this method is 70.0 ms. Considering that the typical action block time window processed by this method is usually more than 1 second, and that retiming calculation and trajectory execution can be performed in parallel (e.g., calculating the retiming trajectory of the next action block during the execution of the previous action block), this overhead is completely within an acceptable range and does not affect real-time control performance. In comparison, the execution efficiency improvement gained by this method (saving about 31 seconds) far outweighs the small delay caused by retiming calculation.

[0066] Furthermore, the comparison results between the baseline method and our method in terms of path advancement rate, velocity envelope, and acceleration envelope are as follows: Figure 2 , Figure 3 , Figure 4 As shown.

[0067] like Figure 2As shown, the baseline method (slow, point-by-point advancement) exhibits a clear step-like nonlinear characteristic in path advancement, with the path advancement rate remaining constant across multiple time periods. This indicates frequent pauses or waiting during the robotic arm's execution, with an overall completion time of approximately 35 seconds. In contrast, the method of this invention (short-time path advancement) increases the path advancement rate from 0 to 1.0 within approximately 4.35 seconds, with a nearly linear and continuous path advancement curve and no stalling phases. This demonstrates that the method of this invention can fully utilize the entire path information for global time optimization, significantly shortening task execution time while satisfying robot physical constraints, eliminating invalid waiting between action blocks, and achieving efficient and smooth online trajectory execution.

[0068] like Figure 3 As shown, the baseline method exhibits a low peak maximum joint velocity (approximately 0.28 rad / s) throughout the execution process, with frequent fluctuations and multiple starts and stops in the velocity curve. This indicates that the robotic arm's movement is slow and discontinuous, failing to fully utilize the robot's dynamic capabilities. In contrast, the maximum joint velocity of the method described in this invention rapidly increases to approximately 2.0 rad / s within a very short time, significantly higher than the baseline method. Furthermore, the velocity curve is smoother and more continuous, rapidly decreasing to zero upon task completion. This demonstrates that the method of this invention can fully utilize the upper limit of joint velocity without exceeding the robot's physical constraints, achieving efficient and smooth motion execution and significantly improving motion execution efficiency.

[0069] like Figure 4 As shown, the maximum joint acceleration of the baseline method remained at approximately 9 rad / s throughout the entire execution process. 2 The high level of acceleration and the flat, unchanging curve indicate that the robotic arm continuously experiences significant acceleration impact during movement, resulting in less smooth motion. In contrast, the maximum joint acceleration of the method described in this invention rapidly increases to approximately 9 rad / s² in the initial stage of movement. 2 The physical constraint upper limit is reached, then rapidly decreases and tends to stabilize, resulting in a smoother, more natural acceleration curve without abrupt changes or sustained high-amplitude oscillations. This demonstrates that the method of this invention can rationally allocate acceleration resources based on the geometry and physical constraints of the entire path, significantly reducing the impact on the robot body while ensuring execution efficiency, thus improving the smoothness of motion and the long-term stability of the system.

[0070] Based on the same inventive concept, such as Figure 5 As shown, this embodiment of the invention also provides a robot trajectory online adaptation and execution system 500 oriented towards action blocks, including: an action block acquisition module 510, a path construction module 520, a path parameter establishment module 530, a trajectory time parameterization module 540, a gripper synchronization module 550, a control sequence generation module 560, and a backtracking execution module 570.

[0071] The action block acquisition module 510 is used to acquire action blocks that contain multiple future path points.

[0072] The path construction module 520 is used to separate the robot arm dimension and gripper dimension from the action block and then splice them with the current robot state to form the original robot arm trajectory and the original gripper trajectory.

[0073] The path parameter establishment module 530 is used to establish path parameters based on the cumulative execution distance between adjacent path points in the original robotic arm trajectory.

[0074] The trajectory time parameterization module 540 is used to perform time parameterization on the original robotic arm trajectory based on path parameters and physical constraints to obtain the retiming robotic arm trajectory.

[0075] The gripper synchronization module 550 is used to synchronously interpolate the original gripper trajectory based on the path parameters corresponding to the retiming robotic arm trajectory, so as to obtain the retiming gripper trajectory.

[0076] The control sequence generation module 560 is used to sample the retiming trajectory of the robotic arm and gripper according to the robot control cycle, and generate the retiming execution trajectory to control the robot to execute.

[0077] The rollback execution module 570 is used to automatically roll back to the point-by-point execution mode of the original action block when time parameterization fails, and execute each original path point in the action block in sequence.

[0078] Based on the same inventive concept, embodiments of the present invention also provide an electronic device, including a memory and one or more processors, wherein the memory is used to store a computer program, and the processor is used to implement the above-described method for online adaptation and execution of robot trajectories oriented towards action blocks when executing the computer program.

[0079] Based on the same inventive concept, embodiments of the present invention also provide a computer-readable storage medium storing a computer program, which, when executed by a computer, implements the above-described method for online adaptation and execution of robot trajectories oriented towards action blocks.

[0080] It should be noted that the robot trajectory online adaptation and execution system, electronic device, and computer-readable storage medium provided in the embodiments of the present invention all belong to the same inventive concept as the robot trajectory online adaptation and execution method for action blocks. The specific implementation process is detailed in the embodiments of the robot trajectory online adaptation and execution method for action blocks, and will not be repeated here.

[0081] The specific embodiments described above illustrate the technical solution and beneficial effects of the present invention in detail. It should be understood that the above description is only the most preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, additions, and equivalent substitutions made within the scope of the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for online adaptation and execution of robot trajectories oriented towards action blocks, characterized in that, Includes the following steps: Obtain an action block containing multiple future path points. Separate the robot arm dimension and gripper dimension from the action block and then concatenate them with the current robot state to form the original robot arm trajectory and the original gripper trajectory. Path parameters are established based on the cumulative execution distance between adjacent path points in the original robotic arm trajectory. Before establishing the path parameters, the original robotic arm trajectory is deduplicated: the starting point and ending point of the original robotic arm trajectory are retained, and the remaining path points are traversed to remove duplicate points that are the same as the most recent retained path point, and near duplicate points whose weighted distance from the most recent retained path point is less than a preset threshold are removed. The original robotic arm trajectory is time-parameterized based on path parameters and physical constraints to obtain the retiming robotic arm trajectory. This includes: determining the path parameter increment between adjacent grid points on the discrete path parameter grid; defining the discrete path parameter velocity square to be solved at each grid point and the discrete path parameter acceleration to be solved between adjacent grid points; establishing discrete dynamic equations of velocity square between adjacent grid points based on the path parameter increment, discrete path parameter velocity square, and discrete path parameter acceleration; transforming the physical constraints of the robotic arm joint velocity and acceleration into inequality constraints on the discrete path parameter velocity square and discrete path parameter acceleration; solving the discrete dynamic equations under the inequality constraints to obtain a sequence of discrete path parameter velocity square values ​​and corresponding acceleration value sequences; calculating the time increment corresponding to each path parameter increment based on the discrete path parameter velocity square value sequence and discrete path parameter acceleration sequence; accumulating the time increments sequentially to obtain the retiming time at each grid point; updating the execution time of each path point of the original robotic arm trajectory based on the retiming time to form the retiming robotic arm trajectory. The original gripper trajectory is synchronously interpolated based on the path parameters corresponding to the re-timing robotic arm trajectory to obtain the re-timing gripper trajectory. This includes: establishing the corresponding robotic arm path parameter position for each state point in the original gripper trajectory, and constructing an interpolation function of the gripper state with respect to the robotic arm path parameters based on the robotic arm path parameter positions; and calculating the gripper state at any time based on the path parameter values ​​of the re-timing robotic arm trajectory using the interpolation function, so that the gripper movement is synchronized with the progress of the robotic arm along the path. The robot arm trajectory and gripper trajectory after retiming are sampled according to the robot control cycle to generate a retiming execution trajectory to control the robot's execution; when time parameterization fails, it automatically reverts to the point-by-point execution mode of the original action block and executes each original path point in the action block in sequence.

2. The method for online adaptation and execution of robot trajectories oriented towards action blocks according to claim 1, characterized in that, The path parameters established based on the cumulative execution distance between adjacent path points in the original robotic arm trajectory include: Calculate the cumulative joint space weighted distance between adjacent path points in the original robotic arm trajectory, or the cumulative end effector movement distance in Cartesian space, or a weighted combination of both, to obtain the monotonically increasing path parameter value corresponding to each path point. This path parameter value represents the cumulative arc length from the trajectory starting point along the geometric path to the current path point.

3. The method for online adaptation and execution of robot trajectories oriented towards action blocks according to claim 1, characterized in that, The retiming execution trajectory includes: the robotic arm joint position sequence, the robotic arm joint velocity sequence, and the gripper state sequence.

4. A robot trajectory online adaptation and execution system oriented towards action blocks, implemented using the robot trajectory online adaptation and execution method oriented towards action blocks as described in any one of claims 1 to 3, characterized in that, include: Action block acquisition module, path construction module, path parameter establishment module, trajectory time parameterization module, gripper synchronization module, control sequence generation module, and rollback execution module; The action block acquisition module is used to acquire action blocks that contain multiple future path points; The path construction module is used to separate the robotic arm dimension and gripper dimension from the action block and then splice them with the current robot state to form the original robotic arm trajectory and the original gripper trajectory. The path parameter establishment module is used to establish path parameters based on the cumulative execution distance between adjacent path points in the original robotic arm trajectory. The trajectory time parameterization module is used to perform time parameterization on the original robotic arm trajectory based on path parameters and physical constraints to obtain the robotic arm trajectory after retiming. The gripper synchronization module is used to synchronously interpolate the original gripper trajectory based on the path parameters corresponding to the retiming robotic arm trajectory, so as to obtain the retiming gripper trajectory. The control sequence generation module is used to sample the retiming trajectory of the robotic arm and gripper according to the robot control cycle, and generate the retiming execution trajectory to control the robot to execute. The rollback execution module is used to automatically roll back to the point-by-point execution mode of the original action block when time parameterization fails, and execute each original path point in the action block in sequence.

5. An electronic device comprising a memory and one or more processors, the memory being used to store a computer program, characterized in that, The processor is used to implement the robot trajectory online adaptation and execution method oriented towards action blocks as described in any one of claims 1 to 3 when executing a computer program.

6. A computer-readable storage medium storing a computer program thereon, characterized in that, When the computer program is executed by a computer, it implements the robot trajectory online adaptation and execution method oriented towards action blocks as described in any one of claims 1 to 3.

Images