Mastering Robot Manipulation with Multimodal Prompts through Pretraining and Multi-task Fine-tuning

TL;DR

This paper introduces a multimodal prompt-based framework for robot manipulation, combining vision and language signals through pretraining and multi-task fine-tuning, achieving state-of-the-art results on VIMA-BENCH.

Contribution

It presents a novel two-stage training pipeline and a multimodal prompt encoder that enhances robot understanding of combined vision and language cues.

Findings

10% success rate improvement on VIMA-BENCH

Effective multimodal understanding and in-context learning demonstrated

State-of-the-art performance achieved

Abstract

Prompt-based learning has been demonstrated as a compelling paradigm contributing to large language models' tremendous success (LLMs). Inspired by their success in language tasks, existing research has leveraged LLMs in embodied instruction following and task planning. In this work, we tackle the problem of training a robot to understand multimodal prompts, interleaving vision signals with text descriptions. This type of task poses a major challenge to robots' capability to understand the interconnection and complementarity between vision and language signals. In this work, we introduce an effective framework that learns a policy to perform robot manipulation with multimodal prompts from multi-task expert trajectories. Our methods consist of a two-stage training pipeline that performs inverse dynamics pretraining and multi-task finetuning. To facilitate multimodal understanding, we design our multimodal prompt encoder by augmenting a pretrained LM with a residual connection to the visual input and model the dependencies among action dimensions. Empirically, we evaluate the efficacy of our method on the VIMA-BENCH and establish a new state-of-the-art (10% improvement in success rate). Moreover, we demonstrate that our model exhibits remarkable in-context learning ability. Project page: \url{https://midas-icml.github.io/}.