TwinBridge: Digital Twin-Driven Policy and Reward Transfer for Real-World Reinforcement Learning in Laboratory Robotics
TwinBridge: Digital Twin-Driven Policy and Reward Transfer for Real-World Reinforcement Learning in Laboratory Robotics
This paper introduces TwinBridge, a digital twin-driven framework that enables online reinforcement learning for precise laboratory manipulation. By combining 3D Gaussian splatting with physics simulation, the system constructs digital twins to provide policy initialization and dense visual rewards. The framework distills supervision into a visual reward model used during real-world interaction to compute dense rewards for human-in-the-loop policy refinement. Experiments across five real-world laboratory tasks show that TwinBridge achieves a final mean autonomous success rate of 98% with at most 25 minutes of real-world interaction per task, demonstrating that dense rewards improve efficiency over sparse rewards and simulation-based initialization accelerates learning.
The study presents TwinBridge, a framework using digital twins for policy and reward transfer in laboratory robotics, reporting a mean autonomous success rate of 98% across five tasks with up to 25 minutes of real-world interaction per task.
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.
- Environment
- Controlled physical settingOpen source S4
- Control
- Reported autonomousOpen source S4
Reported numbers
Reported success rate
98%
Summed sensor-hours
View original evidence
on history to compute dense rewards for human-in-the-loop policy refinement. Across five real-world laboratory tasks, TwinBridge achieves a final mean autonomous success rate of 98% with at most 25 minutes of real-world interaction per task. Experiments show that dense rewards improve interaction efficiency over sparse rewards, and simulation-based policy initialization further accelerates learning.
Open source S4Reported duration
≤25 minutes
View original evidence
on history to compute dense rewards for human-in-the-loop policy refinement. Across five real-world laboratory tasks, TwinBridge achieves a final mean autonomous success rate of 98% with at most 25 minutes of real-world interaction per task. Experiments show that dense rewards improve interaction efficiency over sparse rewards, and simulation-based policy initialization further accelerates learning.
Open source S4
- paper: not_reported
What remains unknown
Not established in the collected evidence: Data origin.
Reported performance applies to the described task. It does not establish general autonomy or deployment readiness.
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
on history to compute dense rewards for human-in-the-loop policy refinement. Across five real-world laboratory tasks, TwinBridge achieves a final mean autonomous success rate of 98% with at most 25 minutes of real-world interaction per task. Experiments show that dense rewards improve interaction efficiency over sparse rewards, and simulation-based policy initialization further accelerates learning.
Open source S4
Source：arXiv Robotics — research abstracts · arxiv.org