Research map / Reinforcement learning / Offline RL
Weighted Policy Constraints for Offline Reinforcement Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use dynamically updated binary advantage weights for policy constraints instead of a constant weight (ablation) | TD3+BC on D4RL gym locomotion-v2 · D4RL gym locomotion-v2 | total normalized score: 712.9 -> 800.7 | Helped |
| Add the Q-learning objective to weighted behavior cloning (ablation) | wBC on D4RL gym locomotion-v2 · D4RL gym locomotion-v2 | total normalized score: 711.4 -> 800.7 | Helped |
| Replace binary advantage weights with exponential advantage weights (ablation) | wPC on D4RL gym locomotion-v2 · D4RL gym locomotion-v2 | total normalized score: 800.7 -> 752.7 | Hurt |
| Shuffle binary advantage weights, removing preference among samples (ablation) | wPC on D4RL gym locomotion-v2 · D4RL gym locomotion-v2 | total normalized score: 800.7 -> 640.2 | Hurt |
| Use exponential advantage weights instead of constant policy-constraint weights (ablation) | TD3+BC on D4RL gym locomotion-v2 · D4RL gym locomotion-v2 | total normalized score: 712.9 -> 752.7 | Helped |
| Use randomized binary weights instead of constant policy-constraint weights (ablation) | TD3+BC on D4RL gym locomotion-v2 · D4RL gym locomotion-v2 | total normalized score: 712.9 -> 640.2 | Hurt |
| Apply dynamically updated binary advantage-weighted constraints (ablation) | TD3+BC on halfcheetah-medium-replay-v2 · halfcheetah-medium-replay-v2 | normalized score: 44.6 -> 48.3 | Helped |
| Apply dynamically updated binary advantage-weighted constraints (ablation) | TD3+BC on hopper-medium-replay-v2 · hopper-medium-replay-v2 | normalized score: 67.4 -> 97.0 | Helped |
| Apply dynamically updated binary advantage-weighted constraints (ablation) | TD3+BC on walker2d-medium-replay-v2 · walker2d-medium-replay-v2 | normalized score: 83.7 -> 89.9 | Helped |
| Apply dynamically updated binary advantage-weighted constraints (ablation) | TD3+BC on walker2d-random-v2 · walker2d-random-v2 | normalized score: 0.9 -> 0.7 | Hurt |
| Use a pre-determined static binary weight rather than dynamically updating weights (ablation) | wPC on walker2d-medium-v2 · walker2d-medium-v2 | performance | Hurt |
| Use randomized binary weights rather than sample-dependent binary advantage weights (ablation) | wPC on D4RL gym locomotion-v2 · medium-expert tasks | normalized score | Hurt |
Closest papers by meaning
- Adaptive Scaling of Policy Constraints for Offline Reinforcement Learning (ICLR 2026 · similarity 0.93)
- A Minimalist Approach to Offline Reinforcement Learning (NeurIPS 2021 · similarity 0.93)
- MOPO: Model-based Offline Policy Optimization (NeurIPS 2020 · similarity 0.92)
- A Clean Slate for Offline Reinforcement Learning (NeurIPS 2025 · similarity 0.91)
- Constrained Latent Action Policies for Model-Based Offline Reinforcement Learning (NeurIPS 2024 · similarity 0.91)
- Constraint-Adaptive Policy Switching for Offline Safe Reinforcement Learning (AAAI 2025 · similarity 0.91)
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.