AGM Framework for Frozen VLA Policies
AGM: Achievement-Grounded Memory for Closed-Loop Agents with Frozen VLA Policies
AGM improves robotic task performance without requiring test-time large-model inference, making it accessible for developers in resource-constrained environments.
AGM converts open-loop execution into a closed loop of execution, verification, and progress for frozen vision-language-action policies.
The framework uses a 2.43M-parameter verification head and achieves results on RoboMME Counting benchmark.
AGM outperforms the strongest memory-augmented baseline by points on average on PickXTimes and BinFill tasks.The framework shows equally decisive gains when implemented on a physical robot.
AGM: Achievement-Grounded Memory for Closed-Loop Agents with Frozen VLA Policies
Frozen vision-language-action (VLA) policies offer broad manipulation skills but execute open-loop action chunks without tracking task progress, so the agent cannot reliably decide whether to continue, retry, or terminate. External memory is a natural remedy, yet it can be harmful when attempted actions are treated as completed progress, turning local execution errors into persistent task-state errors. We propose Achievement-Grounded Memory (AGM), a lightweight closed-loop framework for frozen VLA policies that represents a task as a subgoal sequence with a progress pointer and advances this memory only after the current subgoal is verified by physical evidence. Proprioceptive interaction cues decide when to verify, while coherent point tracking and language-conditioned cross-view comparison, sourced from frozen foundation models through a single 2.43M-parameter verification head, decide what was achieved. AGM thereby converts open-loop execution into a closed loop of execution, verification, and progress, keeping the policy frozen without test-time large-model inference. On the RoboMME Counting benchmark, AGM reaches on PickXTimes and on BinFill, surpassing the strongest memory-augmented baseline by points on average, and the framework yields equally decisive gains on a physical robot. Reliable embodied memory thus depends more on disciplined state updates than on memory capacity.
This report is categorized under AITOP's "Papers" desk. Key entities and technical topics under continuous tracking include AGM, VLA, agents, robots.
Source attribution: curated from arXiv cs.AI (@Hongbo Gao, Zeyu Ni, Xin Wen, Siyu Xu, Ruifeng Li).
