工业机器人定位误差在线自适应补偿

Adaptive Online Compensation for Industrial Robot Positioning Error

受工业机器人本体结构几何及非几何误差因素的影响,机器人执行末端的实际运动轨迹与其理论规划轨迹往往不一致,这严重限制了机器人在加工领域的拓展应用。另外,通过研究发现机器人除在工作空间上定位误差等级存在差异分布外,在服役时间上随着机器人工作性能的退化也会显著恶化其定位精度。为解决该问题,提出了一种基于定长记忆窗增量学习的机器人定位误差在线自适应补偿方法。在该方法中,首先定量分析机器人定位误差与位姿的相关关系,将工作空间划分为多个位姿区块并创建校准样本库,建立了位姿映射模型的自适应优化机制以克服空间中误差等级差异分布的问题;然后设计了定长记忆窗增量学习算法,克服神经网络模型的灾难性遗忘缺陷,并平衡了在线模式下建立机器人新、旧位姿数据映射关系的精度和效率,解决了机器人性能退化加剧定位误差影响位姿映射模型适用性的问题,从而确保算法的补偿精度稳定在目标精度水平线以上;最后,利用St?ubli机器人和UR机器人对所提方法进行了精度在线补偿实验验证。实验结果表明该方法可将St?ubli机器人的定位误差从0.85 mm降至0.13 mm,将UR机器人的定位误差从2.11 mm降至0.17 mm,明显提高了机器人的定位精度,且通过对比发现所提方法的性能明显优于其他同类已发表方法。

Due to the geometric and non-geometric errors of the robot body structure,the actual trajectory of the robot has a big deviation from its nominal trajectory,which seriously limits the application of the robot in machining.Note that the positioning accuracy of the robot will be significantly deteriorated with the degradation of the working performance of the robot during the service time,in addition to the differential distribution of the positioning error levels in the workingspace of the robot.To cope with this problem,an adaptive online compensation method based on fixed-length memory window incremental learning is proposed to compensate the positioning errors of the industrial robot during long-term service.Firstly,the correlation between positioning errors and robot poses is quantitatively studied,and the workspace is divided into several pose blocks and a calibration sample library is created,thus an adaptive optimization mechanism of mapping model is established to address the problem of differential distribution of error levels in workingspace.Secondly,the incremental learning algorithm with fixed-length memory window is designed to overcome the catastrophic forgetting of neural network model and balance the accuracy and efficiency of establishing the mapping relationship between new and old robot pose data in online mode,solving the problem that robot performance degradation aggravates positioning errors and affects the applicability of pose mapping model.Finally,the proposed method is applied to long-term compensation case of Stäubli robot and UR robot,and experimental result shows the proposed method reduces the positioning error of the Stäubli robot from 0.85 mm to 0.13 mm and UR robot from 2.11 mm to 0.17 mm,respectively,outperforming similar methods.