Research map / Reinforcement learning / Policy gradient and actor-critic
Learning to Score Behaviors for Guided Policy Optimization
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 BGPG's alternating optimization instead of particle approximation | Wasserstein trust-region policy optimization · Pendulum, Hopper: Stand, Hopper: Hop, Walker: Stand | wall-clock time: 3720 -> 777; 26908 -> 10817; 23542 -> 12820; 13497 -> 4082 | Helped |
| Replace the KL-based trust region with a Wasserstein behavior-space trust region | BGPG · Six DeepMind Control Suite and Roboschool environments | task performance | Helped |
| Add Wasserstein regularization to evolution strategies | Swimmer · Swimmer | policy performance | Helped |
| Use BGES rather than NSR-ES or NoisyNet-TRPO | Deceptive-reward exploration · Point and quadruped | goal-reaching performance | Helped |
| Use reward-to-go or final-state embeddings instead of the vector of all states (ablation) | BGES · Point | goal-navigation performance | Mixed |
| Use 2 previous policies rather than 1 or 5 (ablation) | BGES with final-state embedding · Point | goal-navigation performance | Helped |
| Replace Wasserstein regularization with Hellinger, Jensen-Shannon, KL, or total-variation regularization (ablation) | Evolution strategies · Swimmer | policy performance | Hurt |
| Remove the behavior-distance regularizer (ablation) | Evolution strategies · Swimmer | policy performance | Hurt |
| Add an expert's reward-to-go embedding as an imitation signal | BGES · Swimmer | policy performance | Helped |
Closest papers by meaning
- Wasserstein Policy Optimization (ICML 2025 · similarity 0.90)
- Iteratively Refined Behavior Regularization for Offline Reinforcement Learning (NeurIPS 2024 · similarity 0.90)
- Behavior-Regularized Diffusion Policy Optimization for Offline Reinforcement Learning (ICML 2025 · similarity 0.89)
- Policy Regularization on Globally Accessible States in Cross-Dynamics Reinforcement Learning (ICML 2025 · similarity 0.88)
- Score Regularized Policy Optimization through Diffusion Behavior (ICLR 2024 · similarity 0.88)
- Adaptive Advantage-Guided Policy Regularization for Offline Reinforcement Learning (ICML 2024 · similarity 0.88)
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.