AGM: Achievement-Grounded Memory for Closed-Loop Agents with Frozen VLA Policies

Sicherheitsmeldung mit Schweregrad noch nicht bewertet. Technische Details im Tab „Originaltext“; empfohlene Schritte in der Checkliste.

Originaltext der Quelle (nicht übersetzt) – die deutsche Zusammenfassung findest du im Tab „Kurzfassung".

arXiv:2608.29537v1 Announce Type: cross Abstract: 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.

Vollständiger Originalartikel ↗

• Betroffenheit im eigenen Stack prüfen: Versionen/Komponenten abgleichen.
• Originalquelle / Hersteller-Advisory lesen: https://arxiv.org/abs/2608.29537
• Verfügbaren Patch oder Workaround einspielen und dokumentieren.