Research map / Reinforcement learning / Imitation learning
Primal Wasserstein 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 |
|---|---|---|---|
| Use PWIL instead of DAC | D4PG imitation learning · Hopper-v2, Ant-v2, HalfCheetah-v2, Humanoid-v2 | environment return | Helped |
| Use PWIL instead of DAC | D4PG imitation learning · Walker2d-v2 | environment return | No clear effect |
| Use PWIL instead of DAC | D4PG imitation learning · Reacher-v2 | environment return | Hurt |
| Use PWIL instead of DAC | imitation learning · HalfCheetah-v2, Ant-v2, Reacher-v2 | sample efficiency | Hurt |
| Use PWIL instead of DAC | imitation learning · Hopper-v2, Walker2d-v2 | sample efficiency | No clear effect |
| Use PWIL instead of DAC | imitation learning · MuJoCo environments except HalfCheetah-v2 and Reacher-v2 | Wasserstein distance | Helped |
| Match state distributions rather than state-action distributions (ablation) | PWIL · MuJoCo locomotion tasks | environment return | Mixed |
| Use unstandardized Euclidean distance instead of standardized Euclidean distance (ablation) | PWIL · MuJoCo locomotion tasks | environment return | Hurt |
| Remove expert-transition replay-buffer prefill (ablation) | PWIL · Walker2d-v2, Ant-v2 | environment return | Hurt |
| Allow expert support holes to have infinite capacity instead of shrinking the support through pop-outs (ablation) | PWIL · Ant-v2, Walker2d-v2, HalfCheetah-v2 | environment return | Hurt |
| Use an offline TCC-learned visual embedding distance as the imitation metric | PWIL with SAC · Door opening task | success rate | Helped |
Closest papers by meaning
- Offline Imitation from Observation via Primal Wasserstein State Occupancy Matching (ICML 2024 · similarity 0.89)
- Reward-free World Models for Online Imitation Learning (ICML 2025 · similarity 0.88)
- Adversarial Imitation Learning via Boosting (ICLR 2024 · similarity 0.88)
- EvIL: Evolution Strategies for Generalisable Imitation Learning (ICML 2024 · similarity 0.88)
- OLLIE: Imitation Learning from Offline Pretraining to Online Finetuning (ICML 2024 · similarity 0.87)
- Imitation Learning from Observation with Automatic Discount Scheduling (ICLR 2024 · 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.