A Robust and Safe Strategy for Robotic Assembly

Yi Liu, Andreas Verleysen, Francis wyffels

Keywords: Reinfocement Learning, Multimodal System, Force Perception

TL;DR: This paper introduces a reinforcement learning strategy based on visual and force perception information, which adds robustness and safety strategies to the basic framework for assembly tasks.

Abstract: In this work we present a reinforcement learning (RL) based approach for enabling a robot to safely perform assembly-type tasks. The proposed strategy involves both grasping and assembly, although our main focus is on the latter. Instead of a pure visual approach, we opt for a combination of force feedback and visual feedback to perceive the shape and direction of the holes. To ensure safe operation, a force-based dynamic safety lock (DSL) is introduced, which limits the pressing force of the robot and prevents emergency stops from being triggered due to excessive force output. Finally, we train and test the strategy with a simulator and build ablation experiments to illustrate the effectiveness of our method. The strategies are independently tested 500 times in the simulator, and we get an 88.57% success rate with a 4mm gap. These models are transferred to the real world and deployed on a real robot. We conducted independent tests and obtained a 79.63% success rate with a 4mm gap.

Confirmation: I am a student and I would like to be considered for the funding award.