RT-SAFE benchmark: agents hit 94.1% task success but only 0.7% safe success in real-time embodied settings
Lianhuiq · x · 2026-10-13
RT-SAFE (arXiv:2610.09294) benchmarks embodied-agent safety under real-time constraints in a simulated urban environment where the world keeps evolving during inference. Across eight VLMs, task completion is high but only 0.7% of episodes in the hardest setting finish without a safety event; real-time execution increases collisions 12.3x versus static evaluation (94.1% vs 91.3% task success), showing that decision latency itself is a safety risk.
Related event: RT-SAFE Benchmark Exposes Real-Time Safety Blind Spot in Embodied AI(5 posts)→
More from Embodied
- Perseverance rover breaks off-world driving record at 28.10 miles — NASA · 2026-10-13
- ASCII Depth-Perception Art With a Mac and RealSense D435 Camera — chrismatthieu · 2026-10-13
- SFGate slams Zuckerberg's new Muse gadget: 'a parasite, don't fall for it' — sfgate · 2026-10-13
- Leaked NVIDIA slides reveal Factory AI blueprint with Nemoclaw agent framework — firstadopter · 2026-10-13
- Berkeley's QF3 makes RL work for flow policies on real robots with one clipped gradient tweak — andrew_n_carr · 2026-10-13
- Google DeepMind Robotics is hiring a pretraining & data quality research scientist in London — bousmalis · 2026-10-13