Tactile-Informed Action Primitives Mitigate Jamming in Dense Clutter

TL;DR

This work proposes the use of two action primitives— burrowing and excavating—that can fluidize the scene to unjam obstacles and enable continued progress and combines the primitives into a closed loop hybrid control strategy using tactile and proprioceptive information.

Abstract

It is difficult for robots to retrieve objects in densely cluttered lateral access scenes with movable objects as jamming against adjacent objects and walls can inhibit progress. We propose the use of two action primitives -- burrowing and excavating -- that can fluidize the scene to un-jam obstacles and enable continued progress. Even when these primitives are implemented in an open loop manner at clock-driven intervals, we observe a decrease in the final distance to the target location. Furthermore, we combine the primitives into a closed loop hybrid control strategy using tactile and proprioceptive information to leverage the advantages of both primitives without being overly disruptive. In doing so, we achieve a 10-fold increase in success rate above the baseline control strategy and significantly improve completion times as compared to the primitives alone or a naive combination of them.

Figures (6)

• Figure 1: a) An end-effector, covered with soft tactile sensors, reaches toward a target in a cupboard with numerous objects. Using motion primitives can prevent object jamming. Inset images show contact forces---circle diameter represents normal force and arrows represent shear forces---on each side of the end-effector. In this case, the contact forces have triggered a clockwise excavate primitive indicated by the red, curved arrow. b) Typical front view available to the robot for a lateral access scene. c) Image of the right side of the end-effector being pinched and corresponding inset force visualization.
• Figure 2: Schematic visualizations of a) straight line control, b) burrowing action primitive and c--e) a sequence showing the progression of a clockwise excavate action primitive.
• Figure 3: a) Pseudo-randomly generated scene layout on a grid with b) corresponding physical setup to enable repeatable comparisons between control strategies. c-d) Example randomized scenes in PyBullet with darker objects corresponding to heavier objects.
• Figure 4: Distributions of a) normalized distance to the goal and b) completion time for the straight line control case and both proposed action primitives on 300 simulated test scenes and 25 physical test scenes. Distance to goal is represented as proportion of scene depth, $d_{scene}$, and completion time as proportion of max duration, $t_{tot}$. **** indicates $p<0.0001$ according to a Wilcoxon signed rank test.
• Figure 5: Contour plots with varying primitive parameters showing performance relative to the baseline straight line control strategy for a) burrow distance to goal, b) burrow completion time, c) excavate distance to goal, and d) excavate completion time. Even in the worst cases, the burrow and excavate control strategies substantially outperform the straight line strategy, with $\approx$ 3-fold improvement on distance to goal and $\approx$ 20% improvement on completion time.