Research map / Reinforcement learning / Visual and representation RL
SafeDec: Constrained Decoding for Safe Autoregressive Generalist Robot Navigation Policies
Parv kapoor, Akila Ganlath, Michael Clifford, Changliu Liu, Sebastian Scherer, Eunsuk Kang
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 hard constrained decoding (HCD) | SPOC, Flare, PoliFormer · phi_avoid, phi_ordered, phi_no_return | STL satisfaction: SPOC phi_avoid 72.0 -> 100.0 | Improved |
| add robustness constrained decoding (RCD) | SPOC, Flare, PoliFormer · phi_avoid, phi_ordered, phi_no_return | STL satisfaction: SPOC phi_avoid 72.0 -> 93.0 | Improved |
| add HCD | SPOC, Flare, PoliFormer · three STL specs | task success rate: SPOC phi_ordered 80.0 -> 70.5 | Hurt |
| add RCD | SPOC, Flare, PoliFormer · three STL specs | task success rate: PoliFormer phi_avoid 82.5 -> 83.5; SPOC phi_avoid 82.5 -> 76.0 | Mixed |
| use HCD instead of Simplex-style filtering | SPOC, Flare, PoliFormer · three STL specs | task success rate: SPOC phi_avoid 72.0 -> 72.5; Flare 78.5 -> 81.0 | Improved |
| use RCD instead of HCD | SPOC, Flare, PoliFormer · three STL specs | task success rate: Flare phi_ordered 73.0 -> 79.5 | Improved |
| add HCD to safe-RL ISA policy | ISA (SafeVLA) · SafetyChores | safety cost: 0.205 -> 0.015 | Improved |
| add RCD to safe-RL ISA policy | ISA (SafeVLA) · SafetyChores | task success rate: 86.5 -> 86.0 | No clear effect |
| add RCD to FLARE | FLARE · SafetyChores | task success rate: 82.0 -> 79.0 | Hurt |
| add HCD to FLARE | FLARE · SafetyChores | task success rate: 82.0 -> 84.0 | Improved |
| inject Gaussian noise into dynamics model (ablation) | RCD and HCD · CHORES | STL satisfaction: RCD Flare 83.0 -> 81.0; HCD SPOC 100.0 -> 96.0 | Hurt |
| increase beta in RCD (PoliFormer) (ablation) | RCD · CHORES PoliFormer | STL satisfaction and success | Mixed |
Closest papers by meaning
- Zero-Shot Trajectory Planning for Signal Temporal Logic Tasks (NeurIPS 2025 · similarity 0.87)
- SafeVLA: Towards Safety Alignment of Vision-Language-Action Model via Constrained Learning (NeurIPS 2025 · similarity 0.86)
- STRAP: Robot Sub-Trajectory Retrieval for Augmented Policy Learning (ICLR 2025 · similarity 0.86)
- Towards Reliable Code-as-Policies: A Neuro-Symbolic Framework for Embodied Task Planning (NeurIPS 2025 · similarity 0.86)
- ActionUNet: Improving Robustness of VLA Models with Efficient Multi-scale Fine-tuning (NeurIPS 2026 · similarity 0.86)
- PACE: Phase-Aware Chunk Execution for Robot Policies with Action Chunking (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.
