As robotic manipulators start performing more daily tasks, striking can be a useful method for transporting objects because it sig nificantly increases the reachable workspace. However, striking methods are underexplored compared to pick-and-place because of the difficulty of modeling and executing striking interactions. In this paper, we develop an algorithm for striking objects so that they stop at a target location. We start with an optimizer in simula tion that solves for the striking velocity given the relative target position, and perform system identification to set the simulation parameters. However, real-world striking with this model does not have high accuracy because it is unclear which parameters should be considered for identification in practice. Therefore, we finally develop a residual learning approach that subsumes all unmodeled differences between the simulation and the real-world environment into action, that is, striking velocity residuals.
We observe that the residual learning model reduces the mean error in stopping position from an average of 9.12 cm to 1.68 cm, or an 81.6% reduction, compared to the baseline system identification approach
@inproceedings{Sinha2025Residual,
title={Targeted Object Striking for a 7-DoF Manipulator: A Residual Learning Approach},
author={Sinha, Priyansh and Chakraborty, Rishin and Brahmbhatt, Samarth and Govindan, Nagamanikandan},
booktitle={Proceedings of AIR '25},
year={2025},
organization={ACM}
}