A Differentiable Recipe for Learning Visual Non-Prehensile Planar Manipulation

TL;DR

This paper introduces a novel differentiable architecture called DLM that combines neural video decoding with contact mechanics priors to improve learning of visual non-prehensile planar manipulation tasks, especially on unseen objects.

Contribution

The work presents a new modular, fully differentiable architecture that integrates deep learning with contact mechanics for robot manipulation from videos, advancing beyond existing methods.

Findings

DLM outperforms learning-only methods on unseen objects and motions.

Combines neural models with contact mechanics priors effectively.

Demonstrates the benefits of differentiable optimization and simulation in manipulation learning.

Abstract

Specifying tasks with videos is a powerful technique towards acquiring novel and general robot skills. However, reasoning over mechanics and dexterous interactions can make it challenging to scale learning contact-rich manipulation. In this work, we focus on the problem of visual non-prehensile planar manipulation: given a video of an object in planar motion, find contact-aware robot actions that reproduce the same object motion. We propose a novel architecture, Differentiable Learning for Manipulation (\ours), that combines video decoding neural models with priors from contact mechanics by leveraging differentiable optimization and finite difference based simulation. Through extensive simulated experiments, we investigate the interplay between traditional model-based techniques and modern deep learning approaches. We find that our modular and fully differentiable architecture performs better than learning-only methods on unseen objects and motions. \url{https://github.com/baceituno/dlm}.