RoboFoundry: System-as-Policy Evolution for Self-Learning Embodied Agents
RoboFoundry treats the robot agent’s whole support system as the thing to improve, not just the model or one module.
The paper says execution traces are turned into validated updates to memory, context, skills, and recovery behavior. Those updates can be task-specific, then promoted when they recur. Its shared semantic interface is meant to let improvements transfer across different robots. The authors report state-of-the-art results on EmbodiedBench, RoboMemArena, LIBERO-PRO, and real-world zero-shot deployments. HF Daily Papers' note
The paper says execution traces are turned into validated updates to memory, context, skills, and recovery behavior. Those updates can be task-specific, then promoted when they recur. Its shared semantic interface is meant to let improvements transfer across different robots. The authors report state-of-the-art results on EmbodiedBench, RoboMemArena, LIBERO-PRO, and real-world zero-shot deployments. HF Daily Papers' note
score 5