Research map / Reinforcement learning / Offline RL
Optimal Transport for Offline Imitation 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 |
|---|---|---|---|
| label offline trajectories using optimal-transport similarity to one expert demonstration, then train IQL | IQL on D4RL locomotion · locomotion-v2-total | aggregate performance: 692.4 -> 685.5 | No clear effect |
| label offline trajectories using optimal-transport similarity to one expert demonstration, then train IQL | IQL on D4RL Antmaze · antmaze-v0-total | aggregate performance: 378.0 -> 384.0 | Helped |
| label offline trajectories using optimal-transport similarity to one expert demonstration, then train IQL | IQL on D4RL Adroit · adroit-v0-total | aggregate performance: 118.1 -> 122.16 | Helped |
| use a uniform transport plan instead of the optimal transport plan (ablation) | OTR+IQL · hopper-medium-replay-v2 | normalized performance: 84.8 -> 69.8 | Hurt |
| use a uniform transport plan instead of the optimal transport plan (ablation) | OTR+IQL · locomotion datasets | normalized performance: 43.3 -> 43.5; 78.7 -> 80.5; 79.4 -> 77.6 | Mixed |
| increase expert demonstrations from K=1 to K=10 (ablation) | OTR+IQL on locomotion · locomotion-v2-total | aggregate performance: 685.5 -> 694.3 | Helped |
| set squashing-function alpha and beta to 1 instead of 5 (ablation) | OTR on locomotion · locomotion datasets | normalized performance: 87.9 -> 86.9; 84.4 -> 71.2; 80.0 -> 75.7 | Mixed |
| use only expert state sequences rather than state-action pairs with PWIL (ablation) | PWIL+IQL · MuJoCo locomotion | normalized performance: 47.5 -> 1.6; 70.4 -> 2.1; 81.9 -> 0.9 | Hurt |
Closest papers by meaning
- Rethinking Optimal Transport in Offline Reinforcement Learning (NeurIPS 2024 · similarity 0.90)
- SEABO: A Simple Search-Based Method for Offline Imitation Learning (ICLR 2024 · similarity 0.89)
- In-Dataset Trajectory Return Regularization for Offline Preference-based Reinforcement Learning (AAAI 2025 · similarity 0.88)
- DiffStitch: Boosting Offline Reinforcement Learning with Diffusion-based Trajectory Stitching (ICML 2024 · similarity 0.88)
- Enhancing Online Reinforcement Learning with Meta-Learned Objective from Offline Data (AAAI 2025 · similarity 0.87)
- Zero-Shot Offline Imitation Learning via Optimal Transport (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.