VIMA: General Robot Manipulation with Multimodal Prompts
This work addresses the problem of fragmented task specification in robotics for researchers and practitioners by proposing a general-purpose approach, though it is incremental in applying prompt-based learning from NLP to robotics.
The paper tackles the challenge of unifying diverse robot manipulation tasks, such as imitation, language instruction, and visual goals, by expressing them with multimodal prompts, and demonstrates that their transformer-based agent VIMA achieves up to 2.9x higher task success rates in zero-shot generalization compared to alternatives.
Prompt-based learning has emerged as a successful paradigm in natural language processing, where a single general-purpose language model can be instructed to perform any task specified by input prompts. Yet task specification in robotics comes in various forms, such as imitating one-shot demonstrations, following language instructions, and reaching visual goals. They are often considered different tasks and tackled by specialized models. We show that a wide spectrum of robot manipulation tasks can be expressed with multimodal prompts, interleaving textual and visual tokens. Accordingly, we develop a new simulation benchmark that consists of thousands of procedurally-generated tabletop tasks with multimodal prompts, 600K+ expert trajectories for imitation learning, and a four-level evaluation protocol for systematic generalization. We design a transformer-based robot agent, VIMA, that processes these prompts and outputs motor actions autoregressively. VIMA features a recipe that achieves strong model scalability and data efficiency. It outperforms alternative designs in the hardest zero-shot generalization setting by up to $2.9\times$ task success rate given the same training data. With $10\times$ less training data, VIMA still performs $2.7\times$ better than the best competing variant. Code and video demos are available at https://vimalabs.github.io/