Research map / Reinforcement learning / Visual and representation RL
Occupancy Reward Shaping: Improving Credit Assignment for Offline Goal-Conditioned Reinforcement Learning
Aravind Venugopal, Jiayu Chen, Xudong Wu, Chongyi Zheng, Benjamin Eysenbach, Jeff Schneider
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 occupancy reward shaping (ORS) to GCIQL | GCIQL · OGBench mean over 13 datasets | success rate (%): 20.0 -> 44.2 | Improved |
| use occupancy-based instead of graph-based reward shaping | Go-Fresh · OGBench mean over 13 datasets | success rate (%): 35.8 -> 44.2 | Improved |
| use occupancy-based reward shaping instead of hierarchical policy | HIQL · OGBench antmaze-giant-navigate | success rate (%): 72 -> 56 | Hurt |
| use ORS instead of Go-Fresh | Go-Fresh · OGBench antmaze-large-explore | success rate (%): 38 -> 22 | Hurt |
| use ORS reward instead of L2 distance reward (ablation) | ORS · OGBench antmaze-giant-navigate | success rate (%): 3 -> 56 | Improved |
| use ORS reward instead of L2 distance reward (ablation) | ORS · OGBench cube-triple-play | success rate (%): 3 -> 37 | Improved |
| add L2 distance reward to GCIQL (ablation) | GCIQL · OGBench cube-triple-play | success rate (%): 7 -> 3 | Hurt |
| learn Q as cumulative sum of state-action reward (vs ORS-s / ORS-r variants) (ablation) | ORS · OGBench scene-play | success rate (%) | Improved |
| lower expectile kappa from 0.9 to 0.6 (ablation) | ORS · OGBench antmaze-giant-navigate | success rate (%): 1 -> 56 | Improved |
| lower expectile kappa from 0.9 to 0.6 (ablation) | ORS · OGBench puzzle-4x4-play | success rate (%): 21 -> 70 | Improved |
| raise expectile kappa from 0.6 to 0.9 (ablation) | ORS · OGBench cube-triple-play | success rate (%): 4 -> 37 | Improved |
| raise expectile kappa from 0.6 to 0.75 (ablation) | ORS · OGBench scene-play | success rate (%): 70 -> 80 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- HIQL — HIQL: Offline Goal-Conditioned RL with Latent States as Actions (NeurIPS 2023) · better on 1 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Score Models for Offline Goal-Conditioned Reinforcement Learning (ICLR 2024 · similarity 0.90)
- Occupancy-based Policy Gradient: Estimation, Convergence, and Optimality (NeurIPS 2024 · similarity 0.88)
- Flattening Hierarchies with Policy Bootstrapping (NeurIPS 2025 · similarity 0.88)
- OGBench: Benchmarking Offline Goal-Conditioned RL (ICLR 2025 · similarity 0.88)
- DISCOVER: Automated Curricula for Sparse-Reward Reinforcement Learning (NeurIPS 2025 · similarity 0.88)
- Action-Dependent Optimality-Preserving Reward Shaping (ICML 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.
