Robotic policies gain long-term memory without retraining
DRAM: Delta-rule Recurrent Associative Memory for Robot Manipulation Policies
RoboticsArtificial Intelligence
Summary
Robotic systems often need to remember past events to act properly, but most can only use recent information or simple memory windows. The authors propose DRAM, a memory module that can be added to existing robotic programs to help them remember longer histories without changing their main design or retraining everything. DRAM uses a special method to update and recall memories efficiently, improving the robot’s ability to perform tasks that depend on past information. Tests show that it works better than other compact memory methods on a variety of robot control systems.
What this means in practice
- •For robotics engineers: Add a memory module to existing robotic control systems to extend their memory without retraining for improved long-horizon tasks.
- •For automation system integrators: Enhance robots used in complex automation by incorporating DRAM to recall historical context and improve manipulation accuracy.
Authors
Xinyu Zhao, Yixiang Shan, Tao Yang, Runyu Lei, Yiming Zhao, Jiaxin Fan, Zongbao Feng, Peng Jia
Abstract
Robotic manipulation is inherently history-dependent, yet most pretrained robotic policies condition on only the current observation or a short temporal window. Equipping such policies with long-term memory remains challenging: existing approaches either feed the backbone multi-frame observation windows, which substantially increase inference cost, or rely on pre-defined semantic features, which limit task generality and may also require the retraining of the backbone to adapt to the memory. We introduce DRAM (Delta-rule Recurrent Associative Memory), a plug-and-play memory module that can be attached to a wide range of pretrained robotic policies, endowing them with long-horizon memory without architectural modification or backbone retraining, requiring only task-specific post-training of the memory module and action expert. DRAM maintains a fixed-size associative memory using gated delta-rule linear attention, with a modified update that incorporates all tokens within each frame in parallel. An architecture-agnostic readout integrates historical context into action prediction across different policy architectures. Experiments show that DRAM consistently improves frozen pretrained policies over short-context baselines and alternative compact memory designs, validating its effectiveness as a fixed-size, post-hoc memory module trained with the backbone frozen.