RT-cache: Efficient Robot Trajectory Retrieval System

This paper introduces RT-cache, a novel trajectorymemory pipeline that accelerates real-world robot inference by leveraging big-data retrieval and learning from experience. While modern Vision-Language-Action (VLA) models can handle diverse robotic tasks, they often incur high per-step inference costs, resulting in significant latency, sometimes minutes per task. In contrast, RT-cache stores a large-scale Memory of previously successful robot trajectories and retrieves relevant multistep motion snippets, drastically reducing inference overhead. By integrating a Memory Builder with a Trajectory Retrieval, we develop an efficient retrieval process that remains tractable even for extremely large datasets. RT-cache flexibly accumulates real-world experiences and replays them whenever the current scene matches past states, adapting quickly to new or unseen environments with only a few additional samples. Experiments on the Open-X Embodiment Dataset and other real-world data demonstrate that RT-cache completes tasks both faster and more successfully than a baseline lacking retrieval, suggesting a practical, data-driven solution for real-time manipulation.