Research map / Reinforcement learning / Imitation learning
Curriculum Offline Imitating Learning
Minghuan Liu, Hanye Zhao, Zhengyu Yang, Jian Shen, Weinan Zhang, Li Zhao, Tie-Yan Liu
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 |
|---|---|---|---|
| Use adaptive neighboring-policy imitation with a return filter instead of direct behavior cloning | Offline imitation on final datasets · hopper-final | return: 1480.4 (800.2) -> 2872.5 (133.9) | Improved |
| Use COIL rather than the strongest listed baseline | Offline learning · halfcheetah-final | return: 10882.0 (1042.7) -> 9328.5 (1940.6) | Hurt |
| Use COIL rather than the strongest listed baseline | Offline learning · hopper-medium | return: 2557.3 -> 3012.0 (332.2) | Improved |
| Use COIL rather than the strongest listed baseline | Offline learning · hopper-medium-replay | return: 3009.6 -> 1333.7 (271.1) | Hurt |
| Use COIL rather than the strongest listed baseline | Offline learning · halfcheetah-medium-expert | return: 9470.3 (4178.9) -> 10535.6 (3334.9) | Improved |
| Use a larger selected-trajectory count N on halfcheetah-medium-expert (ablation) | COIL · halfcheetah-medium-expert | return | Improved |
| Use a medium selected-trajectory count N (4 or 5) instead of a larger batch (ablation) | COIL · halfcheetah-final | return | Improved |
| Remove the return filter (alpha=1.0) (ablation) | COIL · hopper-final | return | Hurt |
| Remove moving-average smoothing from the return filter (alpha=0.0) (ablation) | COIL · hopper-final | return | Hurt |
| Use rapid moving-average filtering (alpha=0.5) (ablation) | COIL · hopper-final | return | Hurt |
| Replace policy-distance-based curriculum with return-ordered BC (ablation) | COIL · final datasets | return | Hurt |
| Replace policy-distance-based curriculum with buffer-shrinking BC (ablation) | COIL · final datasets | return | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- MOPO — MOPO: Model-based Offline Policy Optimization (NeurIPS 2020) · better on 0 of 5 shared results
- BAIL — BAIL: Best-Action Imitation Learning for Batch Deep Reinforcement Learning (NeurIPS 2020) · better on 2 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
- Enhancing Online Reinforcement Learning with Meta-Learned Objective from Offline Data (AAAI 2025 · similarity 0.90)
- Is Behavior Cloning All You Need? Understanding Horizon in Imitation Learning (NeurIPS 2024 · similarity 0.90)
- SEABO: A Simple Search-Based Method for Offline Imitation Learning (ICLR 2024 · similarity 0.90)
- Learning on One Mode: Addressing Multi-modality in Offline Reinforcement Learning (ICLR 2025 · similarity 0.90)
- RLIF: Interactive Imitation Learning as Reinforcement Learning (ICLR 2024 · similarity 0.89)
- Is Value Learning Really the Main Bottleneck in Offline RL? (NeurIPS 2024 · similarity 0.89)
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.
