Perceiver-Actor: A Multi-Task Transformer for Robotic Manipulation

M.H Shridhar, Lucas Manuelli, Dieter Fox

arXiv (Cornell University) · 2022 · 48 citations · 0 references

DOIFull text

Open access

Concepts

TL;DR

Transformers have revolutionized vision and NLP by scaling with large datasets, yet robotic manipulation suffers from limited, expensive data, and voxelized 3‑D observations provide a strong structural prior for efficient 6‑DoF learning. The study asks whether manipulation can still benefit from Transformers with an appropriate formulation, and addresses this by introducing PerAct, a language‑conditioned behavior‑cloning agent for multi‑task 6‑DoF manipulation. PerAct encodes language goals and RGB‑D voxel observations with a Perceiver Transformer, detects the next best voxel action to output discretized actions, and is trained as a single multi‑task Transformer on 18 RLBench tasks (249 variations) and 7 real‑world tasks (18 variations) from only a few demonstrations per task. PerAct significantly outperforms unstructured image‑to‑action agents and 3D ConvNet baselines across a wide range of tabletop tasks.

Abstract

Transformers have revolutionized vision and natural language processing with their ability to scale with large datasets. But in robotic manipulation, data is both limited and expensive. Can manipulation still benefit from Transformers with the right problem formulation? We investigate this question with PerAct, a language-conditioned behavior-cloning agent for multi-task 6-DoF manipulation. PerAct encodes language goals and RGB-D voxel observations with a Perceiver Transformer, and outputs discretized actions by ``detecting the next best voxel action''. Unlike frameworks that operate on 2D images, the voxelized 3D observation and action space provides a strong structural prior for efficiently learning 6-DoF actions. With this formulation, we train a single multi-task Transformer for 18 RLBench tasks (with 249 variations) and 7 real-world tasks (with 18 variations) from just a few demonstrations per task. Our results show that PerAct significantly outperforms unstructured image-to-action agents and 3D ConvNet baselines for a wide range of tabletop tasks.