Comparative Study of Safety Mechanisms | Shahna et al., arXiv:2601.00610
| Method | Train VR | Eval VR | Avg Reward | Guarantee Level |
|---|---|---|---|---|
| Unconstrained | 0.0059 | 0.0000 | -111.3 | None |
| Reward Shaping | 0.0036 | 0.0000 | -113.1 | Probabilistic |
| Lagrangian | 0.0050 | 0.0110 | -106.6 | Soft |
| CBF Filter | 0.0059 | 0.0000 | -111.3 | Practical |
| Shield | 0.0059 | 0.0000 | -111.3 | Strict |