Research map / Reinforcement learning / Imitation learning
OptionGAN: Learning Joint Reward-Policy Options Using Generative Adversarial Inverse Reinforcement Learning
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 joint reward-policy options instead of a single approximator | IRLGAN on Hopper-v1 · Hopper-v1 | average return: 3736.3 $\pm$ 152.4 -> 3641.2 $\pm$ 105.9; 3715.5 $\pm$ 17.6 | No clear effect |
| Use joint reward-policy options instead of a single approximator | IRLGAN on HalfCheetah-v1 · HalfCheetah-v1 | average return: 3212.9 $\pm$ 69.9 -> 3714.7 $\pm$ 87.5; 3616.1 $\pm$ 127.3 | Helped |
| Use joint reward-policy options instead of a single approximator | IRLGAN on Walker2d-v1 · Walker2d-v1 | average return: 4158.7 $\pm$ 247.3 -> 3858.5 $\pm$ 504.9; 4239.3 $\pm$ 314.2 | Mixed |
| Use joint reward-policy options instead of a single approximator | IRLGAN on one-shot Hopper · Hopper (One-Shot) | average return: 2775.1 $\pm$ 203.3 -> 3409.4 $\pm$ 80.8; 3464.0 $\pm$ 67.8 | Helped |
| Use joint reward-policy options instead of a single approximator | IRLGAN on one-shot HalfCheetah · HalfCheetah (One-Shot) | average return: 1296.3 $\pm$ 177.8 -> 1679.0 $\pm$ 284.2; 2219.4 $\pm$ 231.8 | Helped |
| Use joint reward-policy options instead of a single approximator | IRLGAN on one-shot Walker · Walker (One-Shot) | average return: 3229.8 $\pm$ 145.3 -> 3925.3 $\pm$ 138.9; 3769.40 $\pm$ 170.4 | Helped |
| Use joint reward-policy options instead of a single approximator | IRLGAN on HopperSimpleWall · HopperSimpleWall-v0 | average return: 2897.5 $\pm$ 753.5 -> 3140.3 $\pm$ 674.3; 3272.3 $\pm$ 569.0 | No clear effect |
| Use joint reward-policy options instead of a single approximator | IRLGAN on RoboschoolHumanoidFlagrun · RoboschoolHumanoidFlagrun-v1 | average return: 1455.2 $\pm$ 567.6 -> 1868.9 $\pm$ 723.7; 2113.6 $\pm$ 862.9 | Helped |
| Relax the uniform option-distribution penalty (ablation) | OptionGAN with 4 options on HalfCheetah-v1 · HalfCheetah-v1 | learning speed | Helped |
| Force a uniform distribution over options (ablation) | OptionGAN with 4 options · continuous-control tasks | average return | Hurt |
| Replace the mutual-information penalty with entropy regularization (ablation) | OptionGAN · continuous-control tasks | learning stability | Hurt |
| Apply observation normalization (ablation) | IRLGAN and OptionGAN · all environments | performance | No clear effect |
Closest papers by meaning
- Non-Adversarial Inverse Reinforcement Learning via Successor Feature Matching (ICLR 2025 · similarity 0.90)
- Trust Region Reward Optimization and Proximal Inverse Reward Optimization Algorithm (NeurIPS 2025 · similarity 0.89)
- Is Inverse Reinforcement Learning Harder than Standard Reinforcement Learning? A Theoretical Perspective (ICML 2024 · similarity 0.88)
- Rethinking Inverse Reinforcement Learning: from Data Alignment to Task Alignment (NeurIPS 2024 · similarity 0.88)
- A Geometric Perspective on Reward Function Updates in Inverse Reinforcement Learning (NeurIPS 2026 · similarity 0.88)
- Improving Reward Model Generalization from Adversarial Process Enhanced Preferences (ICML 2025 · 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.