Research map / Multi-agent systems and games / Human-AI and goal-conditioned
SafeLab: An Interactive High-Fidelity Benchmark for Embodied Safety in Scientific Robotics
Fengshuo Bai, Yufeng Li, Ruihai Wu, Peishuo Wang, Yuhan Wang, Bernie Hao Zhu, Yuanfei Wang, Tawei Chou, Jing Gao, Runchuan Zhu, Ying Wen, Yaodong Yang and 1 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add bounded residual RL with trajectory-level safety penalties on a frozen IL policy | DP · SafeLab nominal | SSR domain average: 32.5 -> 75.5 (+43.0) | Improved |
| add bounded residual RL on a frozen IL policy | pi0.5 · SafeLab nominal | SSR domain average: 52.9 -> 86.0 (+33.1) | Improved |
| add residual RL under visual/physical/spatial perturbations | DP · SafeLab perturbed | SSR average: 25.6 -> 70.6 | Improved |
| add residual RL under visual/physical/spatial perturbations | pi0.5 · SafeLab perturbed | SSR average: 42.1 -> 79.5 | Improved |
| raise residual scale alpha from 0.10 to 0.30 (ablation) | DP + residual RL · SafeLab Liquid | SSR: 73.2 -> 64.1 | Hurt |
| raise residual scale alpha from 0.10 to 0.20 (ablation) | DP + residual RL · SafeLab Spatial | SSR: 78.9 -> 81.3 | Improved |
| lower residual scale alpha from 0.10 to 0.05 (ablation) | DP + residual RL · SafeLab Liquid | SSR: 73.2 -> 68.4 | Hurt |
| use constrained RL variants instead of penalty reward (ablation) | residual RL · SafeLab Liquid | SSR / training cost: 1.8-2.1x training cost | No clear effect |
| larger injected base-policy perturbation (small to large) (ablation) | DP + residual RL, alpha 0.1 · Recovery under injected perturbation | recovery rate: 88% -> 17% | Hurt |
| evaluate with safety (SSR) instead of task success (SR) | VLA/IL policies · SafeLab all domains | success rate gap: SSR trails SR by more than 30 points | Hurt |
Closest papers by meaning
- AGENTSAFE: Benchmarking the Safety of Embodied Agents on Hazardous Instructions (CVPR 2026 · similarity 0.90)
- IS-Bench: Evaluating Interactive Safety of VLM-Driven Embodied Agents in Daily Household Tasks (AAAI 2026 · similarity 0.89)
- LabUtopia: High-Fidelity Simulation and Hierarchical Benchmark for Scientific Embodied Agents (NeurIPS 2025 · similarity 0.88)
- Sim-to-Lab-to-Real: Safe Reinforcement Learning with Shielding and Generalization Guarantees (Abstract Reprint) (AAAI 2024 · similarity 0.86)
- CaP-X: A Framework for Benchmarking and Improving Coding Agents for Robot Manipulation (ICML 2026 · similarity 0.86)
- SENTINEL: A Multi-Level Formal Framework for Safety Evaluation of Foundation Model-based Embodied Agents (NeurIPS 2026 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
