The authors propose FAITH, which learns a state-action safety value function using a safety critic and a backup actor, then amortizes minimal-intervention filtering with a feedforward network.
Safety filter separates task and safety learning in robot control
FAITH trains a task policy through a learned feedforward safety filter, avoiding competing objectives and handling infeasible states by minimizing predicted harm.
Big Tech
Songyuan Zhang · Baljeet Singh · Sarthak Ranjeet Kaingade · Chuchu Fan · Bryan Trinh
Massachusetts Institute of Technology · Amazon
Research Digest··3 min read
Zhang et al.
Why this paper
From Amazon and Massachusetts Institute of Technology
In one line
FAITH trains a robot's task policy through a learned safety filter, achieving high safety and performance without competing objectives.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors (2 noted)
- ✓Reports numbers on named benchmarks
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§