REA: A Routed Embodied Agent Runtime for Asynchronous Brain-Cerebellum Control
REA: A Routed Embodied Agent Runtime for Asynchronous Brain-Cerebellum Control
This research presents REA (Routed Embodied Agent), a unified runtime system that enables asynchronous coordination between deliberative foundation models and fast reactive policies. The system employs a route memory, boundary-triggered judge gates, and shared-budget escalation to prevent slow brain queries from stalling execution. Evaluated on 3,000 test episodes across 60 LIBERO-Pro tasks, REA achieves 95.9% success, outperforming the best single-agent baseline (94.7%) while reducing control steps by 25% and control time by 17%. In a separate study on RoboTwin 2.0, REA reaches 84% success with VLA policies and 86% with VLA+code, compared to 80% for the strongest single VLA baseline (MEDex). The approach also transfers to BEHAVIOR-1K and a real Unitree G1 humanoid robot.
The paper introduces REA, a runtime architecture designed to coordinate slow deliberative foundation models with fast reactive policies without stalling execution, evaluated on LIBERO-Pro and RoboTwin benchmarks.
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.
Reported numbers
test episodes
3,000
Reported trials
View original evidence
escalation ladder for bounded recovery. Evaluated on 3,000 test episodes across 60 LIBERO-Pro tasks, REA reaches 95.9% success and beats the best single-agent baseline in its own policy pool (94.7%) while reducing control steps by 25% and control time by 17%. In a separate 100-episode RoboTwin 2.0 study, REA achieves 84% success with VLA policies and 86% with VLA+code, compared with 80% for the strongest single VLA
Open source S5LIBERO-Pro tasks
60
View original evidence
escalation ladder for bounded recovery. Evaluated on 3,000 test episodes across 60 LIBERO-Pro tasks, REA reaches 95.9% success and beats the best single-agent baseline in its own policy pool (94.7%) while reducing control steps by 25% and control time by 17%. In a separate 100-episode RoboTwin 2.0 study, REA achieves 84% success with VLA policies and 86% with VLA+code, compared with 80% for the strongest single VLA
Open source S5
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
escalation ladder for bounded recovery. Evaluated on 3,000 test episodes across 60 LIBERO-Pro tasks, REA reaches 95.9% success and beats the best single-agent baseline in its own policy pool (94.7%) while reducing control steps by 25% and control time by 17%. In a separate 100-episode RoboTwin 2.0 study, REA achieves 84% success with VLA policies and 86% with VLA+code, compared with 80% for the strongest single VLA
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org