Research map / Reinforcement learning / Imitation learning
CEIL: Generalized Contextual Imitation Learning
Jinxin Liu, Li He, Yachen Kang, Zifeng Zhuang, Donglin Wang, Huazhe Xu
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 CEIL instead of the strongest listed baseline | Offline single-domain LfD · D4RL medium/medium-replay/medium-expert, 4 environments | sum of normalized scores: 754.6 -> 1163.5 | Improved |
| Use CEIL instead of the strongest listed baseline | Offline single-domain LfO · D4RL medium/medium-replay/medium-expert, 4 environments | sum of normalized scores: 741.7 -> 821.7 | Improved |
| Use CEIL instead of the strongest listed baseline | Offline cross-domain LfD · D4RL medium/medium-replay/medium-expert, 4 environments | sum of normalized scores: 749.7 -> 780.9 | Improved |
| Use CEIL instead of the strongest listed baseline | Offline cross-domain LfO · D4RL medium/medium-replay/medium-expert, 4 environments | sum of normalized scores: 724.4 -> 727.3 | Improved |
| Add expert-matching optimization of the hindsight embedding function f_phi (ablation) | CEIL · S-off-LfD, D4RL medium/medium-replay/medium-expert | sum of normalized scores: 1030.8 -> 1163.5 | Improved |
| Add the J_MI latent-space expert-matching objective (ablation) | CEIL · S-off-LfD, D4RL medium/medium-replay/medium-expert | sum of normalized scores: 951.4 -> 1163.5 | Improved |
| Add expert-matching optimization of the hindsight embedding function f_phi (ablation) | CEIL · S-off-LfO, D4RL medium/medium-replay/medium-expert | sum of normalized scores: 759.2 -> 821.7 | Improved |
| Add the J_MI latent-space expert-matching objective (ablation) | CEIL · S-off-LfO, D4RL medium/medium-replay/medium-expert | sum of normalized scores: 792.0 -> 821.7 | Improved |
| Add cross-domain regularization that encourages task-relevant embeddings (ablation) | CEIL · Cross-domain offline IL | normalized score | Improved |
| Add expert observations and cross-domain expert behaviors to the offline training data | CEIL · Hybrid offline IL, Hopper-v2 | normalized score: 29.4 -> 58.6 | Improved |
| Use CEIL for online learning from observations | CEIL · Online LfO, Walker2d-v2 | return | Hurt |
Later papers that compared with this method
1 later accepted paper reports CEIL as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Auto-regressive In-context Demonstration Selection (ICML 2026 · better on 9 of 9 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
- DualCOIL: Offline Imitation Learning from Contrasting Demonstrations (ICML 2026 · similarity 0.89)
- Instant Policy: In-Context Imitation Learning via Graph Diffusion (ICLR 2025 · similarity 0.88)
- OLLIE: Imitation Learning from Offline Pretraining to Online Finetuning (ICML 2024 · similarity 0.88)
- Seeing Differently, Acting Similarly: Heterogeneously Observable Imitation Learning (ICLR 2023 · similarity 0.88)
- Is Behavior Cloning All You Need? Understanding Horizon in Imitation Learning (NeurIPS 2024 · similarity 0.88)
- SEABO: A Simple Search-Based Method for Offline Imitation Learning (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.
