Novel Peg-in-Hole Robot Assembly via Misalignment-Error Estimation

Recent studies on autonomous robotic manipulation for manufacturing automation have heavily relied on environmental perception through vision, based on artificial intelligence. However, vision-recognition errors are inevitable, and misalignment can lead to jamming, decreasing the task success rates, and potentially damaging workpieces. This letter introduces a peg-in-hole control approach that merges a parallel position/force controller with a network, designed to estimate misalignment errors. In real-world experiments, we confirmed that, under poor gain-tuning conditions, the success rate improved by up to 44% compared with conventional methods, even without additional parameter tuning. By effectively preventing jamming caused by misalignment in peg-in-hole tasks, the proposed method enhances robotic-assembly performance in unstructured environments.