Driving · 2026-10-05
RoboBridge: A Self-Evolving Embodied Agent Framework for Sim-to-Real Transfer
Chenxi Li, Zhangrui Zhao, Rui Li, Yuan Gao, Kehui Liu, Jiarui Li, Dong Wang, Tong Si, Minting Pan, Wanli Ouyang, Dongzhan Zhou
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
RoboBridge treats sim-to-real transfer as continual adaptation of explicit, executable task skills. It stores task knowledge as procedures linking intent, observations, tool operations, and outcome verification. Interaction feedback generates candidate skill revisions that are validated before persistence or rejection. It uses a frozen pretrained vision-language-action policy as a reusable action tool, adding inference-time reward guidance without updating policy parameters, and evaluates transfer on LIBERO-PRO and corresponding real-robot tasks.
Why it matters
The authors target sim-to-real transfer where end-to-end vision-language-action policies often require calibration, demonstrations, and further policy training. RoboBridge decouples reusable procedural task knowledge from frozen action capabilities, aiming to preserve transferable skill structure while selectively revising environment-dependent operations using rollout feedback and validation-gated updates.
Method
- Represents task knowledge as editable procedural skills connecting task intent, observations, tool operations, and outcome verification; skills are revised via embodied execution feedback.
- Implements a feedback-driven skill evolution loop where candidate revisions are evaluated through task execution and retained only if they remain valid and improve performance; rejected updates are excluded from persistent knowledge.
- Uses a pretrained end-to-end VLA policy as a frozen action tool within an agent harness; when needed, the agent configures inference-time reward guidance to steer denoising while keeping policy parameters fixed.
Abstract (from arXiv)
A key challenge in bringing embodied intelligence into the real world is transferring capabilities from simulation to reality and enabling agents to continually adapt after deployment. End-to-end vision-language-action policies provide strong manipulation capabilities, but their transfer to physical environments typically relies on calibrating simulated visual and dynamical conditions, collecting additional target-domain demonstrations, and optimizing the policy through further training. Tool-using embodied agents offer flexible task orchestration, yet existing systems primarily emphasize task execution and experience reuse within a given environment, with limited support for transferring procedural knowledge and continuously adapting it across simulation and reality. We propose RoboBridge, a framework that treats sim-to-real transfer as the continued adaptation of executable task skills. The agent represents task knowledge as procedures connecting task intent, observations, tool operations, and outcome verification. Interaction feedback is used to generate candidate skill revisions, which are evaluated before being persisted or rejected. A pretrained vision-language-action policy is exposed as a reusable action tool and enhanced with inference-time guidance, enabling fine-grained execution without retraining the underlying policy. RoboBridge grounds transferable skills in task semantics and interaction interfaces shared across simulation and reality. This representation preserves reusable task structure while allowing environment-dependent operations to be selectively revised through real-world execution feedback. We evaluate the framework on LIBERO-PRO and corresponding physical tasks, studying both skill evolution and post-transfer adaptation. Our framework provides a route from one-shot policy deployment to continual procedural learning across environments.
Related papers
- Explore, Execute, Evolve: A Skill Acquisition and Reuse Loop for Embodied Agents
- EmbodiedEvo: Self-improving Coding Agents for Physical Intelligence
- Skill2Real: Agentic Skill Learning for Zero-Shot Sim-to-Real Robot Manipulation
- RobotWorld: Benchmarking Multimodal Agents for Robot Use Across Diverse Tasks and Embodiments
- EmbodiedCLBench: Evaluating Continual Learning for Self-Evolving Embodied Agents
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-10-02 · Indexed by Pub-AI: 2026-10-05 UTC
RoboBridge: A Self-Evolving Embodied Agent Framework for Sim-to-Real Transfer
Preprint posted: 2026-10-02 · Original record ↗
- arxiv
- 2610.02717
Explicit source versions (1)
- v1 · 2026-10-02
- Preprint available
Source date: 2026-10-02 · First detected: 2026-10-25 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
A key challenge in bringing embodied intelligence into the real world is transferring capabilities from simulation to reality and enabling agents to continually adapt after deployment. End-to-end vision-language-action policies provide strong manipulation capabilities, but their transfer to physical environments typically relies on calibrating simulated visual and dynamical conditions, collecting additional target-domain demonstrations, and optimizing the policy through further training. Tool-using embodied agents offer flexible task orchestration, yet existing systems primarily emphasize task execution and experience reuse within a given environment, with limited support for transferring procedural knowledge and continuously adapting it across simulation and reality. We propose RoboBridge, a framework that treats sim-to-real transfer as the continued adaptation of executable task skills. The agent represents task knowledge as procedures connecting task intent, observations, tool operations, and outcome verification. Interaction feedback is used to generate candidate skill revisions, which are evaluated before being persisted or rejected. A pretrained vision-language-action policy is exposed as a reusable action tool and enhanced with inference-time guidance, enabling fine-grained execution without retraining the underlying policy. RoboBridge grounds transferable skills in task semantics and interaction interfaces shared across simulation and reality. This representation preserves reusable task structure while allowing environment-dependent operations to be selectively revised through real-world execution feedback. We evaluate the framework on LIBERO-PRO and corresponding physical tasks, studying both skill evolution and post-transfer adaptation. Our framework provides a route from one-shot policy deployment to continual procedural learning across environments.