Untangling Dense Non-Planar Knots by Learning Manipulation Features and Recovery Policies

Abstract

Robot manipulation for untangling 1D deformable structures such as ropes, cables, and wires is challenging due to their infinite dimensional configuration space, complex dynamics, and tendency to self-occlude. Analytical controllers often fail in the presence of dense configurations, due to the difficulty of grasping between adjacent cable segments. We present two algorithms that enhance robust cable untangling, LOKI and SPi-DERMan, which operate alongside HULK, a high-level planner from prior work. LOKI uses a learned model of manipulation features to refine a coarse grasp keypoint prediction to a precise, optimized location and orientation, while SPiDERMan uses a learned model to sense task progress and apply recovery actions. We evaluate these algorithms in physical cable untangling experiments with 336 knots and over 1500 actions on real cables using the da Vinci surgical robot. We find that the combination of HULK, LOKI, and SPiDERMan is able to untangle dense overhand, figure-eight, double-overhand, square, bowline, granny, stevedore, and triple-overhand knots. The composition of these methods successfully untangles a cable from a dense initial configuration in 68.3% of 60 physical experiments and achieves 50% higher success rates than baselines from prior work. Supplementary material, code, and videos can be found at https://tinyurl.com/rssuntangling.

Disentangling Dense Multi-Cable Knots

“…Prior work has studied the task of single-cable untangling from both loose [13] and dense [6,22] initial configurations, where dense configurations lack space between crossings. Lui et al [13] propose modeling a cable configuration via a graphical abstraction representing cable crossings and endpoints, and approximate this model from RGB-D input through analytical feature engineering.…”

“…HULK also does not apply out-of-the-box to the multi-cable setting due to ambiguity in selecting knot-loosening actions when there are many cable endpoints from which to trace under-crossings. Lastly, Sundaresan et al [22] define single-cable algorithms LOKI (Local Oriented Knot Inspection) and SPiDERMan (Sensing Progress in Dense Entanglements for Recovery Manipulation), which plan grasp refinement steps such as recentering the knot in the workspace and removing the knot from the gripper jaws when it becomes wedged there. Similarly to HULK, LOKI and SPiDERMan must be modified for the multi-cable scenario to account for crossings containing >3 cable segments and crossings involving cables of similar or different colors.…”

“…2: Graph Representation: Provided a dense initial square knot (A), we take a Node Deletion move specified by hold (green) and pull (dark blue) keypoints, yielding a looser configuration shown in (B). We use a graphical abstraction to model the state of intertwined cables, extending previous work on modelling single-cable configurations [6,13,22]. In this graph, endpoints and intra/inter-cable crossings constitute nodes, and edges denote over (+) and under (-) crossings, shown in (C).…”

“…Previous works define a graph structure to concisely represent the configuration space of single-cable knots [6,13,22] and introduce algorithms that operate on this compressed representation to perform untangling. This work extends these models to represent n-cable knots.…”

“…To implement IRON-MAN from visual observations, we employ keypoint regression methods [6] to learn pull and hold keypoints. We plan manipulation actions over the learned keypoints with five manipulation primitives: two from knot theory and prior work [6,13], two from Sundaresan et al [22], and a novel move introduced in this work.…”