2AM: Grounding Agent-Side Memory as Guidance for Steerable Action Models in Long-Horizon Manipulation
2AM: Grounding Agent-Side Memory as Guidance for Steerable Action Models in Long-Horizon Manipulation
This research paper introduces 2AM, a framework that assigns task memory to a multimodal agent while using an RGB-based, stateless action model for motion execution. By compiling interaction history into subtask language and hints, the system steers the action model without relying on depth sensors, online geometry, or external planners. Trained with structured hint labels and robustness augmentations like condition dropout and spatial noise, 2AM achieves a 76.3% average completion rate on the LIBERO-Mem benchmark, representing a 61.5-point improvement over the strongest baseline of 14.8%, with 63.0% relaxed and 11.8% strict success rates.
The study isolates the contribution of agent-side memory to long-horizon manipulation by constraining tool breadth and training an RGB-based action model under noise, achieving a 76.3% average completion rate on LIBERO-Mem without depth or online geometry.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
e and optional 2D grasp, place, and move hints that bind its physical intention at different time scales. To teach this steerability to the VLA, we augment demonstrations with structured hint labels and train under condition dropout, spatial noise, and temporal jitter to tolerate imperfect Agent outputs. On LIBERO-Mem, without depth, online geometry, or planner-based object motion, 2AM reaches 76.3% average completio
Open source S4
Source：arXiv Robotics — research abstracts · arxiv.org