MemER: Scaling Up Memory for Robot Control via Experience Retrieval
This addresses the challenge of memory-efficient robot control for long-horizon manipulation tasks, representing an incremental improvement by integrating memory retrieval into existing vision-language-action models.
The paper tackles the problem of enabling robot policies to use memory for long-horizon tasks by proposing a hierarchical framework where a high-level policy selects relevant keyframes from experience to guide a low-level policy, and it outperforms prior methods on three real-world robotic manipulation tasks requiring minutes of memory.
Humans routinely rely on memory to perform tasks, yet most robot policies lack this capability; our goal is to endow robot policies with the same ability. Naively conditioning on long observation histories is computationally expensive and brittle under covariate shift, while indiscriminate subsampling of history leads to irrelevant or redundant information. We propose a hierarchical policy framework, where the high-level policy is trained to select and track previous relevant keyframes from its experience. The high-level policy uses selected keyframes and the most recent frames when generating text instructions for a low-level policy to execute. This design is compatible with existing vision-language-action (VLA) models and enables the system to efficiently reason over long-horizon dependencies. In our experiments, we finetune Qwen2.5-VL-7B-Instruct and $π_{0.5}$ as the high-level and low-level policies respectively, using demonstrations supplemented with minimal language annotations. Our approach, MemER, outperforms prior methods on three real-world long-horizon robotic manipulation tasks that require minutes of memory. Videos and code can be found at https://jen-pan.github.io/memer/.