ContextVLA: Vision-Language-Action Model with Amortized Multi-Frame Context
This work addresses efficiency and performance issues in robotic behavior cloning for partially observable environments, representing an incremental improvement over existing multi-frame methods.
The paper tackles the problem of inconsistent performance gains when using multi-frame observations in partially observable robotic tasks by introducing ContextVLA, a Vision-Language-Action model that compresses past observations into a single context token, resulting in improved task performance with reduced training and inference times compared to single-frame models.
Leveraging temporal context is crucial for success in partially observable robotic tasks. However, prior work in behavior cloning has demonstrated inconsistent performance gains when using multi-frame observations. In this paper, we introduce ContextVLA, a policy model that robustly improves robotic task performance by effectively leveraging multi-frame observations. Our approach is motivated by the key observation that Vision-Language-Action models (VLA), i.e., policy models built upon a Vision-Language Model (VLM), more effectively utilize multi-frame observations for action generation. This suggests that VLMs' inherent temporal understanding capability enables them to extract more meaningful context from multi-frame observations. However, the high dimensionality of video inputs introduces significant computational overhead, making VLA training and inference inefficient. To address this, ContextVLA compresses past observations into a single context token, allowing the policy to efficiently leverage temporal context for action generation. Our experiments show that ContextVLA consistently improves over single-frame VLAs and achieves the benefits of full multi-frame training but with reduced training and inference times.