A robot program was optimized with SPI-DP to place a cup at four target poses on two shelf levels, using approach motions based on human demonstrations while avoiding collisions. All trials were collision-free, and the target pose was reached with a mean accuracy of 0.6 mm.
The robot program for the cup experiment was optimized again for a wine glass and a changed gripper geometry, additionally optimizing the target pose of the transfer skill to minimize cycle time. SPI-DP optimization reduced motion length by 2.8 cm on average and decreased cycle time by 40% while ensuring collision-free movements.
This experiment evaluates the scalability of joint motion and task-level optimization in complex industrial robot programs using a poka-yoke quality assurance task. A UR5 robot arm performs force-controlled searches to locate holes on an engine block, with randomized offsets to simulate process noise. SPI-DP optimizes the robot's motions and task parameters, improving cycle time and success probability. After optimization, hole detection rates increased significantly, with a 62% reduction in search duration, and motion-level optimization ensured collision-free movements.
Hole 2 faster since directly missed
Hole 2 slower but found successfully
This paper presents Shadow Program Inversion with Differentiable Planning (SPI-DP), a novel first-order opti- mizer capable of optimizing robot programs with respect to both high-level task objectives and motion-level constraints. To that end, we introduce Differentiable Gaussian Process Motion Planning for N-DoF Manipulators (dGPMP2-ND), a differentiable collision-free motion planner for serial N-DoF kinematics, and integrate it into an iterative, gradient-based optimization approach for generic, parameterized robot pro- gram representations. SPI-DP allows first-order optimization of planned trajectories and program parameters with respect to objectives such as cycle time or smoothness subject to e.g. collision constraints, while enabling humans to understand, modify or even certify the optimized programs. We provide a comprehensive evaluation on two practical household and industrial applications.
@article{alt2024spidp,
title={Shadow Program Inversion with Differentiable Planning: A Framework for Unified Robot Program Parameter and Trajectory Optimization},
author={Benjamin Alt and Claudius Kienle and Darko Katic and Rainer Jäkel and Michael Beetz},
year={2024},
eprint={2409.08678},
archivePrefix={arXiv},
primaryClass={cs.RO},
url={https://arxiv.org/abs/2409.08678},
}
This work was supported by the German Federal Ministry of Education and Research (grant 01MJ22003B), the DFG CRC EASE (CRC #1320) and the EU project euROBIN (grant 101070596).