Visual and Force-Driven-Based Assembly Learning Using Collaborative Robots

Abstract

This thesis explores learning frameworks for collaborative robots in assembly operations. Reinforcement Learning (RL) with visual and haptic information tackles target uncertainty, while proactive actions improve policy learning. Another framework combines visual servoing-based Learning from Demonstration (LfD) and force-based Learning by Exploration (LbE) for efficient programming. Lastly, a sim-to-real transfer learning framework addresses sample efficiency and safety concerns, using CycleGAN and force control transfer for successful real-world adaptation.