SCAPE: Scenario-Conditioned Simulation-Augmented Policy Evaluation
Research
Source: arXiv cs.ROPublish time unverified
arXiv:2608.19425v1 Announce Type: new Abstract: Reliable performance evaluation is a central bottleneck for deploying robot-learning policies in real-world conditions. Real-world testing is faithful but costly and difficult to scale, whereas simulation-based testing scales easily but is inevitably biased by the sim-to-real gap.