RoboFoundry: System-as-Policy Evolution for Self-Learning Embodied Agents
RoboFoundry evolves embodied agent systems as policies, lifting GPT-5.5 by 27.8% on EmbodiedBench.
RoboFoundry is an embodied-agent framework that treats the supporting system as a self-evolving policy, converting execution traces into validated updates of context memory and hierarchical skills. A shared semantic interface separates embodiment-invariant decisions from robot-specific execution so improvements can transfer across robots. On EmbodiedBench it improves GPT-5.5 by 27.8% and brings Qwen3.7-Plus to 70.3% versus 72.7% for GPT-5.5. It also leads RoboMemArena by at least 39.0% and beats Cap-Agent0 by 243.8%–679.7% across LIBERO-PRO perturbations, with zero-shot real-robot transfer.
- Treats memory, skills, and context as one evolving system policy
- Raises GPT-5.5 by 27.8% on EmbodiedBench
- Qwen3.7-Plus reaches 70.3% versus GPT-5.5 at 72.7%
- Beats all RoboMemArena baselines by at least 39%
- Outperforms Cap-Agent0 by 243.8% to 679.7% on LIBERO-PRO
Full article238 words · extracted from huggingface.co · click to collapse
A foundation model should not act in isolation as an embodied agent. Yet, existing methods often optimize individual components of the agent stack, such as memory, context, skills, or action interfaces, rather than treating the supporting system itself as a unified policy. Moreover, interaction alone does not yield self-improvement unless execution experience is converted into persistent, validated system changes. We therefore propose RoboFoundry, the first embodied agentic framework that formulates this process as Self-Evolving System-as-Policy. RoboFoundry diagnoses capability gaps in decision-making and memory management, converts execution traces into validated task-specific system updates, and promotes recurring improvements to the general system. Evolution operates over two complementary surfaces: a context system that manages active internal context and persistent file-system memory, and a hierarchical skill system that organizes atomic skills, reusable compositions, and failure-conditioned recovery. A shared semantic interface separates embodiment-invariant decisions from embodiment-specific execution, allowing evolved system capabilities to transfer across heterogeneous robots. On EmbodiedBench, RoboFoundry achieves state-of-the-art performance, notably improving GPT-5.5 by 27.8%. It also brings Qwen3.7-Plus to near parity with GPT-5.5 (70.3% vs. 72.7%), showing consistent gains from system-as-policy evolution across foundation models. For long-horizon memory, RoboFoundry outperforms all baselines on RoboMemArena by at least 39.0%, even against methods assisted by external foundation models. On LIBERO-PRO, it further outperforms Cap-Agent0 by 243.8%-679.7% across all perturbation types. In real-world deployments, RoboFoundry demonstrates zero-shot transfer and online evolution across robots and tasks, highlighting its potential for fully autonomous embodied agents.
Text extracted automatically; images, tables and formatting may be missing. Original: https://huggingface.co/papers/2609.32862