Impact-Aware Control using Time-Invariant Reference Spreading

TL;DR

It is shown how a nonsmooth physics engine can be used to construct this impact model for complex scenarios, which warrants applicability to a large range of possible impact states without requiring contact stiffness and damping parameters.

Abstract

With the goal of increasing the speed and efficiency in robotic manipulation, a control approach is presented that aims to utilize intentional simultaneous impacts to its advantage. This approach exploits the concept of the time-invariant reference spreading framework, in which partly-overlapping ante- and post-impact reference vector fields are used. These vector fields are coupled via an impact model in proximity of the expected impact area, minimizing the otherwise large impact-induced velocity errors and control efforts. We show how a nonsmooth physics engine can be used to construct this impact model for complex scenarios, which warrants applicability to a large range of possible impact states without requiring contact stiffness and damping parameters. In addition, a novel interim-impact control mode provides robustness in the execution against the inevitable lack of exact impact simultaneity and the corresponding unreliable velocity error during the time when contact is only partially established. This interim mode uses a position feedback signal that is derived from the ante-impact velocity reference to promote contact completion, and smoothly transitions into the post-impact mode. An experimental validation of time-invariant reference spreading control is presented for the first time through a set of 600 robotic hit-and-push and dual-arm grabbing experiments.

Figures (10)

• Figure 1: Depiction of the dual-arm robotic setup used for the experimental validation of the control approach presented in this work.
• Figure 2: Ante- and post-impact velocity reference vector fields depicted for a dual-arm 3DOF planar use case: (a) shows the ante-impact velocity reference field for robot 1 (blue) and robot 2 (red) together with the nominal end effector path (green); (b) shows the post-impact velocity reference field with $r_\text{min}^p = 0$ (yellow) and the two velocity fields that together comprise this reference field (blue and red). The blue field represents the term that corresponds to the predicted post-impact velocity $\bm v^+_{o,\text{est}}$, and the red field represents the term that eventually ensures convergence to the point $\bm p_{o,f}$; (c) shows the nominal path followed by the center of the object when the velocity reference field is followed in yellow, versus the red path taken when the red vector field from (b) is followed without accounting for the predicted post-impact velocity.
• Figure 3: Visualization of the simulation environment for a dual-arm grabbing use case.
• Figure 4: Snapshots of the system for one of the hit-and-push experiments. In (a), the selected initial configuration is shown. In (b), it is shown that the ante-impact velocity field is followed to align the end effector with the impact direction before impacting the object as shown in (c). After following a velocity field that is adjusted to the predicted impact map, (d) shows the final configuration of the system after the object is pushed to its desired location.
• Figure 5: Snapshots of the system for one of the dual-arm grabbing experiments. In (a), the initial configuration is shown, with no symmetry between the two robots with respect to the object. In (b), it is shown that the robots synchronize while following the ante-impact velocity field to create a simultaneous impact between the two robots and the object. Due to an unmodeled 15mm displacement of the object, the right robot impacts the object first in (c), which triggers the interim mode. In (d), it is shown that the impact sequence is eventually successfully resolved, with (e) then showing that the path followed by tracking the post-impact velocity reference matches the direction of the predicted post-impact velocity. In (f), it is shown that the same post-impact velocity reference eventually steers the object to a desired post-impact rest pose.