Vision-Based High-Precision Assembly with Force Feedback

Assembling objects in unstructured scenarios re-quires accurate 3D position estimation, which presents a sig-nificant challenge in achieving high-precision manipulation. This paper proposes a method to reduce positioning errors in high-precision assembly problems by combining visual and force feedback. Our approach leverages the strengths of both sensors by analyzing visual positioning characteristics and using force feedback to locate the target position during inaccuracies. We evaluate our approach through experiments of inserting a nut onto a screw without threading, conducted in both simulation and real-world scenarios. Our results demonstrate an improvement in the assembly success rate compared to previous methods.