Research map / Reinforcement learning / Visual and representation RL
Generating Behaviorally Diverse Policies with Latent Diffusion Models
Shashank Hegde, Sumeet Batra, K.R. Zentner, Gaurav S. Sukhatme
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 |
|---|---|---|---|
| increase GHN decoder size from GHN8 to GHN32 (ablation) | latent diffusion · Humanoid | QD ratio: 0.77 -> 0.94 | Improved |
| increase GHN decoder size from GHN8 to GHN32 (ablation) | latent diffusion · Humanoid | mean error in measure: 0.267 -> 0.129 | Improved |
| increase GHN decoder size from GHN8 to GHN32 (ablation) | latent diffusion · Humanoid | compression ratio: 19:1 -> 8:1 | Hurt |
| raise KL coefficient well above 1e-6 (ablation) | VAE · reward ratio and JS divergence | measure divergence | Hurt |
| filter out text labels containing fall (ablation) | language conditioned diffusion · sequential composition | success rate (%): 38 -> 59 | Improved |
| sample only text labels containing quickly (ablation) | language conditioned diffusion · sequential composition | success rate (%): 38 -> 79 | Improved |
| language instead of measure conditioning for sequences | latent diffusion · sequential composition | success rate (%): 80 -> 38 | Hurt |
| compare reconstructed with original archive on Ant | latent diffusion · Ant | coverage (%): 71.03 -> 52.66 | Hurt |
| compare reconstructed with original archive on Halfcheetah | latent diffusion · Halfcheetah | QD score (x10^7): 11.4 -> 11.1 | No clear effect |
Closest papers by meaning
- Score Regularized Policy Optimization through Diffusion Behavior (ICLR 2024 · similarity 0.89)
- GenPO: Generative Diffusion Models Meet On-Policy Reinforcement Learning (NeurIPS 2025 · similarity 0.88)
- Quality Diversity through Human Feedback: Towards Open-Ended Diversity-Driven Optimization (ICML 2024 · similarity 0.88)
- Behavioral Foundation Models for Quality Diversity (NeurIPS 2026 · similarity 0.88)
- Proximal Policy Gradient Arborescence for Quality Diversity Reinforcement Learning (ICLR 2024 · similarity 0.88)
- Training Diffusion Models Towards Diverse Image Generation with Reinforcement Learning (CVPR 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.
