Cables are commonplace in homes, hospitals, and industrial warehouses and are prone to tangling. This paper extends prior work on autonomously untangling long cables by introducing novel uncertainty quantification metrics and actions that interact with the cable to reduce perception uncertainty. We present Sliding and Grasping for Tangle Manipulation 2.0 (SGTM 2.0), a system that autonomously untangles cables approximately 3 meters in length with a bilateral robot using estimates of uncertainty at each step to inform actions. By interactively reducing uncertainty, Sliding and Grasping for Tangle Manipulation 2.0 (SGTM 2.0) reduces the number of state-reseting moves it must take, significantly speeding up run-time. Experiments suggest that SGTM 2.0 can achieve 83% untangling success on cables with 1 or 2 overhand and figure-8 knots, and 70% termination detection success across these configurations, outperforming SGTM 1.0 by 43% in untangling accuracy and 200% in full rollout speed.

Endpoint & Knot Detection

We use Faster R-CNN models to detect knots and endpoints in the cable.

SGTM 2.0 first detects the number of knots and endpoints in the scene. If the endpoints are not visible, there is no way to verify any knot's relative position to the endpoint. This is necessary because SGTM 2.0 only untangles knots adjacent to an endpoint to avoid knots colliding into each other and creating irrecoverable configurations. If fewer than two endpoints and no knots are visible, the algorithm is also unable to perform a termination check as that requires performing an incremental Reidemeister, which grasps the cable at the endpoints. In both these cases, SGTM 2.0 performs an endpoint exposure. If two endpoints are visible and no knots are visible, SGTM 2.0 proceeds to the incremental Reidemeister move. If one or two endpoints are visible and there are knots in the scene, it attempts to untangle, beginning by tracing from the visible endpoint(s). Here, if it is not able to confidently trace from either endpoint to a knot, SGTM 2.0 performs a Reidemeister move or endpoint exposure (based on the number of endpoints visible) to increase likelihood of unambiguous traces in future steps. Otherwise, it assesses the cage-pinch network uncertainty on the predicted points. If it is confident, it proceeds with a full cage-pinch dilation. Else, it performs a partial cage-pinch dilation to disambiguate the state.