Planning-Query-Guided Model Generation for Model-Based Deformable Object Manipulation

TL;DR

This paper presents a diffusion-based method for automatically generating task-specific, spatially adaptive dynamics models to improve planning efficiency in deformable object manipulation tasks.

Contribution

It introduces a novel model generator that predicts per-region model resolutions based on planning queries, enhancing planning speed with minimal performance loss.

Findings

Doubles planning speed on a tree-manipulation task

Achieves small decrease in task performance with adaptive models

Demonstrates potential for using prior data to generate efficient models

Abstract

Efficient planning in high-dimensional spaces, such as those involving deformable objects, requires computationally tractable yet sufficiently expressive dynamics models. This paper introduces a method that automatically generates task-specific, spatially adaptive dynamics models by learning which regions of the object require high-resolution modeling to achieve good task performance for a given planning query. Task performance depends on the complex interplay between the dynamics model, world dynamics, control, and task requirements. Our proposed diffusion-based model generator predicts per-region model resolutions based on start and goal pointclouds that define the planning query. To efficiently collect the data for learning this mapping, a two-stage process optimizes resolution using predictive dynamics as a prior before directly optimizing using closed-loop performance. On a tree-manipulation task, our method doubles planning speed with only a small decrease in task performance over using a full-resolution model. This approach informs a path towards using previous planning and control data to generate computationally efficient yet sufficiently expressive dynamics models for new tasks.