Research map / Reinforcement learning / Policy gradient and actor-critic
Training and Evaluation of Deep Policies Using Reinforcement Learning and Generative Models
Ali Ghadirzadeh, Petra Poklukar, Karol Arndt, Chelsea Finn, Ville Kyrki, Danica Kragic, Mårten Björkman
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 |
|---|---|---|---|
| add a low beta (0.001) to the VAE generative model (ablation) | GenRL · Hockey task | max reward: -0.267 -> -0.040 | Improved |
| use VAE instead of InfoGAN as generative model (ablation) | GenRL · Basketball task | max reward: -0.081 vs -0.469 | Improved |
| use lambda 3.5 (stronger disentanglement) in InfoGAN (ablation) | GenRL · Basketball task | max reward: -1.183, -1.019, -1.344 | Hurt |
| higher recall of generative model (ablation) | GenRL · Hockey task (VAE models) | Pearson R with policy reward: 0.959 | Improved |
| higher disentanglement of generative model (ablation) | GenRL · Hockey task (VAE models) | Pearson R with policy reward: 0.979 | Improved |
| higher precision of generative model (ablation) | GenRL · Hockey task (all models) | Pearson R with policy reward: -0.068 | No clear effect |
| higher recall of generative model (ablation) | GenRL · Basketball task (all models) | Pearson R with policy reward: 0.907 | Improved |
| use GenRL instead of PPO and SAC | GenRL · Hockey and basketball tasks | average reward during training | Improved |
| use GenRL instead of PPO and SAC | GenRL · Hockey task safety | table-top collisions per episode | Improved |
Closest papers by meaning
- GenFlowRL: Shaping Rewards with Generative Object-Centric Flow in Visual Reinforcement Learning (ICCV 2025 · similarity 0.90)
- GenPO: Generative Diffusion Models Meet On-Policy Reinforcement Learning (NeurIPS 2025 · similarity 0.89)
- Deep Reinforcement Learning for Robotics: A Survey of Real-World Successes (AAAI 2025 · similarity 0.89)
- Foundation Policies with Hilbert Representations (ICML 2024 · similarity 0.88)
- RoboGen: Towards Unleashing Infinite Data for Automated Robot Learning via Generative Simulation (ICML 2024 · similarity 0.87)
- ESRL: Efficient Sampling-Based Reinforcement Learning for Sequence Generation (AAAI 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.
