Research map / Reinforcement learning / Imitation learning
SMILe: Scalable Meta Inverse Reinforcement Learning through Context-Conditional Policies
Seyed Kamyar Seyed Ghasemipour, Shixiang (Shane) Gu, Richard Zemel
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 max-ent IRL (SMILe state-action) instead of behavioral cloning (Meta-BC) | context-conditional meta-imitation policy · HalfCheetah Random Velocity, 4 demos 20 sub | abs velocity error (lower better): 0.931 -> 0.181 | Improved |
| use max-ent IRL (SMILe state-action) instead of behavioral cloning (Meta-BC) | context-conditional meta-imitation policy · HalfCheetah Random Velocity, 64 demos 20 sub | abs velocity error (lower better): 0.468 -> 0.102 | Improved |
| use max-ent IRL (SMILe state-action) instead of behavioral cloning (Meta-BC) | context-conditional meta-imitation policy · Walker random dynamics | custom normalized metric: -0.403 -> 0.710 | Improved |
| use SMILe instead of Meta-DAgger (online expert BC) | context-conditional meta-imitation policy · Walker random dynamics | custom normalized metric: 0.275 -> 0.710 | Improved |
| use SMILe instead of Meta-DAgger (online expert BC) | context-conditional meta-imitation policy · HalfCheetah Random Velocity, 64 demos 1 sub | abs velocity error (lower better): 0.082 -> 0.118 | Hurt |
| use SMILe instead of Meta-BC | context-conditional meta-imitation policy · Ant Linear Classification | success rate: 0.50 -> 0.68 | Improved |
| use state-only demonstrations instead of state-action (ablation) | SMILe · HalfCheetah Random Velocity, 64 demos 1 sub | abs velocity error (lower better): 0.118 -> 0.138 | No clear effect |
| increase number of expert demonstrations per task from 4 to 64 (ablation) | Meta-BC · HalfCheetah Random Velocity | abs velocity error (lower better): 0.931 -> 0.468 | Improved |
Closest papers by meaning
- Hybrid Inverse Reinforcement Learning (ICML 2024 · similarity 0.91)
- Is Inverse Reinforcement Learning Harder than Standard Reinforcement Learning? A Theoretical Perspective (ICML 2024 · similarity 0.90)
- Inverse Reinforcement Learning by Estimating Expertise of Demonstrators (AAAI 2025 · similarity 0.90)
- Recent Advancements in Inverse Reinforcement Learning (AAAI 2024 · similarity 0.89)
- Simplifying Constraint Inference with Inverse Reinforcement Learning (NeurIPS 2024 · similarity 0.89)
- Imitating Language via Scalable Inverse Reinforcement Learning (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.
