漸進式視覺伺服於六自由度跨物體插件組裝任務

Abstract

插件組裝是一項具有挑戰性的機器操作任務，因為它對估計誤差的容忍度很低。先前的方法依賴於力-扭矩控制或端到端視覺伺服，難以實現六自由度插入任務和處理較大的初始對準誤差。此外，在沒有進行高成本的重新訓練下，缺乏泛化能力使他們無法處理沒看過的目標。為此，我們提出漸進式視覺伺服（CFVS）插件組裝方法，該方法利用三維點雲訊息實現了能插入任意傾斜角的六自由度插件組裝。此外，CFVS 能夠透過快速姿態估計來處理較大的初始對準誤差，然後再經過細化調整。再者，通過引入置信度熱圖，CFVS 對各種形狀的目標都具有穩健性。大量實驗表明 CFVS 優於最先進的方法，在 3-DoF、4-DoF 和 6-DoF 插件中的平均成功率分別為 100%、91% 和 82%。

Peg-in-hole is a challenging robotic manipulation task due to its low tolerance against estimation error. Prior methods rely on force-torque control or end-to-end visual servoing, having difficulty achieving 6-DoF insertion and handling large initial alignment errors. Moreover, the lack of generalization ability prevents them from dealing with unseen target objects without costly re-training. To this end, we propose a Coarse-to-Fine Visual Servoing (CFVS) peg-in-hole assembly method, which first achieves the 6-DoF peg-in-hole assembly with arbitrary tilt angle by exploiting 3D point-cloud information. Also, CFVS is capable of handling large initial alignment errors through a fast pose estimation before refinement. Furthermore, by introducing a confidence heatmap, CFVS is robust against various shapes of targets. Extensive experiments show that CFVS outperforms state-of-the-art methods and obtains 100%, 91%, and 82% average success rates in 3-DoF, 4-DoF, and 6-DoF peg-in-hole, respectively.