Research map / Reinforcement learning / Offline RL
VOCE: Variational Optimization with Conservative Estimation for Offline Safe Reinforcement Learning
Jiayi Guan, Guang Chen, Jiaming Ji, Long Yang, Ao Zhou, Zhijun Li, changjun jiang
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 |
|---|---|---|---|
| remove conservative lower-bound Q estimation for rewards (VOCE-Qr) (ablation) | VOCE · Point/Car Goal/Button | episode reward | Hurt |
| remove conservative upper-bound Q estimation for costs (VOCE-Qc) (ablation) | VOCE · Point/Car Goal/Button | episode cost | Hurt |
| remove conservative cost estimation (VOCE-Qc) (ablation) | VOCE · Point/Car Goal/Button | episode reward | No clear effect |
| raise fraction phi of constraint-satisfying trajectories to 0.7-0.9 (ablation) | VOCE · Point-Goal, Point-Button | reward and cost | Improved |
| raise phi to 1.0 (all trajectories safe) (ablation) | VOCE · Point-Goal, Point-Button | reward and cost | Hurt |
| use VOCE instead of BCQ-Lag, C-CRR, COptiDICE | VOCE · Goal and Button tasks | episode reward | Improved |
| use VOCE on Goal tasks across phi values | VOCE · Point-Goal, Car-Goal | episode cost | Improved |
| use VOCE on Button tasks with small phi | VOCE · Point-Button, Car-Button | episode cost | Mixed |
| use VOCE instead of C-CRR for training time per step | VOCE · Point-Button | training time (s): 0.0891 -> 3.1498 | Hurt |
| use VOCE instead of C-CRR for test-time per step | VOCE · Point-Button | testing time (s): 1.5140 × 1e-4 -> 3.0665 × 1e-4 | No clear effect |
Closest papers by meaning
- Latent Safety-Constrained Policy Approach for Safe Offline Reinforcement Learning (ICLR 2025 · similarity 0.90)
- POCE: Primal Policy Optimization with Conservative Estimation for Multi-constraint Offline Reinforcement Learning (CVPR 2024 · similarity 0.89)
- Online Optimization for Offline Safe Reinforcement Learning (NeurIPS 2025 · similarity 0.88)
- Enhancing Efficiency of Safe Reinforcement Learning via Sample Manipulation (NeurIPS 2024 · similarity 0.88)
- A Provable Approach for End-to-End Safe Reinforcement Learning (NeurIPS 2025 · similarity 0.87)
- Offline Safe Reinforcement Learning Using Trajectory Classification (AAAI 2025 · similarity 0.87)
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.
