High Performance Robot Trajectory Tracking through Iterative Learning and Optimization

As industrial robots find increasing applications that demand precise path tracking, such as spraying and welding, the need for optimal performance and productivity becomes paramount. However, achieving a balance between tracking accuracy, path speed, and motion uniformity can present conflicting objectives. The current programming process involves motion primitives consisting of waypoints connected by pre-defined motion segments, where users specify parameters like path speed and blending zone. However, the actual robot motion relies on the robot joint servo controller and joint motion constraints, which remain largely unknown to users.

The process of programming a robot to achieve submillimeter tracking accuracy is currently time-consuming and predominantly manual, involving the intricate tuning of numerous coupled parameters in the motion primitives. The performance outcome is also influenced by additional choices, including possible redundant degrees of freedom, the location of the target curve, and the robot configuration. The complexity is further heightened when multiple robots are involved.

To address these challenges, we have developed a systematic approach to generate robot motion for spatial curve tracking and optimize robot motion parameters. The process commences with first optimizing static parameters, including redundant degrees of freedom resolution, robot and part placement, and robot pose selection. It then progresses to selecting suitable motion primitives and iteratively updating waypoints to minimize the tracking error. The iterative update of the waypoints draws on reinforcement learning and approximate gradient descent algorithms. The ultimate performance objective is to maximize path speed while adhering to tracking accuracy and speed uniformity constraints throughout the entire path.

The effectiveness of this approach has been successfully demonstrated through both simulation and physical implementations on ABB and FANUC robots, tackling two challenging example curves, with single and multiple robots. Comparing the results with the baseline using the current industry practice, the optimized performance exhibits over 100% improvement.

This new robot motion optimization approach significantly advances the current state of industrial robot motion planning, enabling precision and efficiency in complex spatial curve tracking tasks. By systematically optimizing robot motion parameters, this approach paves the way for enhanced productivity and performance in critical industrial applications.