Research map / Reinforcement learning / Imitation learning
Imitation by Predicting Observations
Andrew Jaegle, Yury Sulsky, Arun Ahuja, Jake Bruce, Rob Fergus, Greg Wayne
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 observation-transition generative reward models instead of an adversarial discriminator | Imitation from observations · DeepMind Control Suite, 13 tasks | return | Mixed |
| Use FORM rather than GAIfO+GP under task-irrelevant distractors | Imitation from observations · DeepMind Control Suite with N=8 or N=16 distractor dimensions | return | Improved |
| Use autoregressive model-rollout regularization with β_ar=1.0 rather than β_ar=0.1 (ablation) | FORM · Humanoid run | return: 668.4 ± 19.4 -> 694.0 ± 10.8 | Improved |
| Vary autoregressive noise weight (ablation) | FORM · Quadruped run | return: 952.7 ± 3.6; 945.4 ± 2.4; 955.5 ± 0.4 | Mixed |
| Increase maximum overshooting offset from 1 to 3 (ablation) | FORM · Humanoid run | return: 266.1 ± 190.5 -> 645.9 ± 32.1 | Improved |
| Vary maximum overshooting offset (ablation) | FORM · Quadruped run | return: 920.57 ± 11.6; 920.15 ± 1.2; 921.52 ± 9.8 | No clear effect |
| Increase L2 regularization weight from 0.1 to 1.0 (ablation) | FORM · Humanoid run | return: 694.0 ± 10.8 -> 602.5 ± 74.3 | Hurt |
| Vary L2 regularization weight (ablation) | FORM · Quadruped run | return: 945.4 ± 2.4; 953.0 ± 2.8; 918.4 ± 0.7 | Mixed |
| Condition GAIfO on two frames rather than one (ablation) | GAIfO · DeepMind Control Suite | return | Mixed |
Closest papers by meaning
- Imitation Learning from Observation with Automatic Discount Scheduling (ICLR 2024 · similarity 0.91)
- Imitation Learning by Estimating Expertise of Demonstrators (ICML 2022 · similarity 0.91)
- Imitation Learning by Reinforcement Learning (ICLR 2022 · similarity 0.91)
- EvIL: Evolution Strategies for Generalisable Imitation Learning (ICML 2024 · similarity 0.90)
- Is Behavior Cloning All You Need? Understanding Horizon in Imitation Learning (NeurIPS 2024 · similarity 0.90)
- Interactive and Hybrid Imitation Learning: Provably Beating Behavior Cloning (NeurIPS 2025 · 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.
