This technical note discloses our implementation of a six degree-of-freedom (DOF) high-precision robotic phantom on a commercially available industrial robot manipulator. These manipulators are designed to optimize their set point tracking accuracy as it is the most important performance metric for industrial manipulators. Their in-house controllers are tuned to suppress its error less than a few tens of micrometers. However, the use of industrial robot manipulators in six DOF robotic phantom can be a difficult problem since their in-house controller are not optimized for continuous path tracking in general. Although instantaneous tracking error in a continuous path tracking task will not exceed five millimeters during motion with the in-house controller, it seriously matters for a robotic phantom, as the tracking error should remain within one millimeter in three dimensional space for all time during motion. The difficulty of the task is further increased since the reference trajectory of a robotic phantom, which is a six DOF tumor motion of a patient, cannot be as smooth as the ones used in factories. The present study presents a feedforward controller for a feedback-controlled industrial six DOF robotic manipulator to be used as a six DOF robotic phantom to drive the water equivalent phantom (WEP). We first trained a set of six recurrent neural networks (RNNs) to capture the six DOF input/output behavior of the robotic manipulator controlled by its in-house controller, and we proceed to formulate an iterative learning control (ILC) using the trained model to generate an augmented reference trajectory for a specific patient that enables very high tracking accuracy to that trajectory. Experimental evaluation results demonstrate clear improvements in the accuracy of the proposed robotic phantom compared to our previous robotic phantom, which uses the same manipulator but is driven by a different corrected reference trajectory.

本技术说明公开了我们在商用工业机器人机械手上实现六自由度 (DOF) 高精度机器人模型。这些机械手旨在优化其设定点跟踪精度，因为它是工业机械手最重要的性能指标。他们的内部控制器经过调整，可将其误差抑制在几十微米以内。然而，在六自由度机器人模型中使用工业机器人机械手可能是一个难题，因为它们的内部控制器通常没有针对连续路径跟踪进行优化。尽管在使用内部控制器进行运动期间，连续路径跟踪任务中的瞬时跟踪误差不会超过 5 毫米，但对于机器人模型来说，它非常重要，因为在运动过程中，跟踪误差应在三维空间中始终保持在一毫米以内。由于机器人模型的参考轨迹（患者的六自由度肿瘤运动）不能像工厂使用的那样平滑，因此任务的难度进一步增加。本研究提出了一个前馈控制器，用于反馈控制的工业六自由度机器人机械手，用作六自由度机器人体模来驱动水当量体模 (WEP)。我们首先训练了一组六个循环神经网络 (RNN) 来捕捉由其内部控制器控制的机器人机械手的六个自由度输入/输出行为，我们继续使用经过训练的模型制定迭代学习控制 (ILC)，为特定患者生成增强的参考轨迹，从而实现对该轨迹的非常高的跟踪精度。实验评估结果表明，与我们之前的机器人体模相比，所提出的机器人体模的准确性明显提高，该机器人体模使用相同的机械手，但由不同的校正参考轨迹驱动。