Research map / Reinforcement learning / Imitation learning
Random Expert Distillation: Imitation Learning via Expert Policy Support Estimation
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 RED's fixed support-estimation reward instead of GAIL | TRPO imitation learning · Hopper | episodic reward: 3614.22 $\pm$ 7.17 -> 3625.96 $\pm$ 4.32 | Helped |
| Use RED's fixed support-estimation reward instead of GAIL | TRPO imitation learning · HalfCheetah | episodic reward: 4515.70 $\pm$ 549.49 -> 3072.04 $\pm$ 84.71 | Hurt |
| Use RED's fixed support-estimation reward instead of GAIL | TRPO imitation learning · Reacher | episodic reward: -32.37 $\pm$ 39.81 -> -10.43 $\pm$ 5.20 | Helped |
| Use RED's fixed support-estimation reward instead of GAIL | TRPO imitation learning · Walker2d | episodic reward: 4877.98 $\pm$ 2848.37 -> 4481.37 $\pm$ 20.94 | Hurt |
| Use RED's fixed support-estimation reward instead of AE | TRPO imitation learning · Ant | episodic reward: 3778.61 $\pm$ 422.63 -> 3552.77 $\pm$ 348.67 | Hurt |
| Replace RND with an overparametrized, L2-regularized autoencoder for support estimation (ablation) | RED-style imitation learning · Mujoco tasks | episodic reward | Mixed |
| Add BC initialization rather than starting from a random policy (ablation) | RED · HalfCheetah and Ant | episodic reward | Helped |
| Add the terminal reward heuristic (ablation) | RED and AE · Autonomous driving | distance travelled without collision | Helped |
| Use fixed rewards from support estimation instead of dynamically updated adversarial rewards | RED versus GAIL · Simple Domain | mean episodic reward | Helped |
| Use fixed rewards from support estimation instead of moment-matching rewards | RED versus GMMIL · Simple Domain | mean episodic reward | Helped |
Closest papers by meaning
- Imitation Learning by Reinforcement Learning (ICLR 2022 · similarity 0.94)
- Generative Adversarial Imitation Learning (NeurIPS 2016 · similarity 0.93)
- Hybrid Inverse Reinforcement Learning (ICML 2024 · similarity 0.90)
- EvIL: Evolution Strategies for Generalisable Imitation Learning (ICML 2024 · similarity 0.90)
- Off-Dynamics Reinforcement Learning via Domain Adaptation and Reward Augmented Imitation (NeurIPS 2024 · similarity 0.90)
- Rethinking Inverse Reinforcement Learning: from Data Alignment to Task Alignment (NeurIPS 2024 · similarity 0.90)
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.