Push-T

Table of Contents

Evolution of VoSI profile over a rollout

Task Summary

Push-T is a stylized 2D non-prehensile manipulation task that involves complex contact dynamics between a cylindrical end-effector (projected to a circle) and a T-shaped object that must be pushed into a target configuration.

Visualization of MixL trajectories

These visualizations help illustrate how the VoSI metric is realized by comparing the difference in performance of mixed loop (MixL) execution strategies (that forgo sensory information for different horizons h and subsequently operate closed loop) from a closed loop execution strategy (denoted as CL). The open loop phase of the rollouts are represented with a gray indicator on the top-right, which the closed loop phase represented with a green indicator.

0.mp4

1.mp4

Media 1(a-b). From states close to the reset distribution we observe that there is no degradation util the end-effector makes first contact with the T and pushes it for some duration. Over extended open loop actions over the push phase of the task there is degradation in performance and delay in potentially correctly aligning the T in the desired configuration.

2.mp4

Media 1(c). From a state where the T is already aligned with the desired target position -- we observe that is no serious degradation in performance but given that the open-loop actions synthesized by TD-MPC2 policy have the tendency to return close to the T which can potentially perturb the achieved configuration.

Evolution of VoSI profile over a rollout

vosi_cl_rollout.mp4

Media 2. Visualization of the VoSI profiles over the course of a close loop execution starting from the reset distribution. Observe that some amount of degradation sets in when committing to extended duration of open loop actions when the end-effector is in contact with the T. Once the configuration is achieved there is no additional value gain from sensing.

Page updated

Google Sites

Report abuse