Research map / Reinforcement learning / Policy gradient and actor-critic
Efficient Wasserstein Natural Gradients for 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 |
|---|---|---|---|
| Add WNG to BGPG and combine it with the behavioral Wasserstein penalty | Policy gradients · RS Walker2d | average return: 652.53 -> 739.51 {\pm 81.10} | Helped |
| Add WNG to BGPG and combine it with the behavioral Wasserstein penalty | Policy gradients · RS HalfCheetah | average return: 1668.75* -> 1817.68 \pm 79.64 | Helped |
| Add WNG to BGPG and combine it with the behavioral Wasserstein penalty | Policy gradients · RS Ant | average return: 1735.98 -> 2072.56 \pm 85.24 | Helped |
| Add WNG to BGPG and combine it with the behavioral Wasserstein penalty | Policy gradients · RS Reacher | average return: 0.46 -> 4.37 \pm 2.23 | Helped |
| Add WNG to BGPG and combine it with the behavioral Wasserstein penalty | Policy gradients · DoublePendulum | average return: 6526.48 -> 7254.98 \pm 419.36 | Helped |
| Add WNG to BGPG and combine it with the behavioral Wasserstein penalty | Policy gradients · Hopper | average return: 1934.43 -> 2439.56 \pm 394.39 | Helped |
| Replace the BGPG update with a WNG update | Policy gradients · RS Walker2d | average return: 652.53 -> 683.38^{\star} | Helped |
| Replace the BGPG update with a WNG update | Policy gradients · RS HalfCheetah | average return: 1668.75* -> 1655.04 | No clear effect |
| Replace the BGPG update with a WNG update | Policy gradients · RS Ant | average return: 1735.98 -> 1844.14* | Helped |
| Replace the BGPG update with a WNG update | Policy gradients · RS Reacher | average return: 0.46 -> 2.13* | Helped |
| Replace the BGPG update with a WNG update | Policy gradients · DoublePendulum | average return: 6526.48 -> 6740.96* | Helped |
| Replace the BGPG update with a WNG update | Policy gradients · Hopper | average return: 1934.43 -> 2235.23* | Helped |
Closest papers by meaning
- Policy Optimization as Wasserstein Gradient Flows (ICML 2018 · similarity 0.92)
- Learning to Score Behaviors for Guided Policy Optimization (ICML 2020 · similarity 0.89)
- Wasserstein Policy Optimization (ICML 2025 · similarity 0.89)
- Distributional Reinforcement Learning with Regularized Wasserstein Loss (NeurIPS 2024 · similarity 0.87)
- Natural Policy Gradient as Doubly Smoothed Policy Iteration: A Bellman-Operator Framework (NeurIPS 2026 · similarity 0.87)
- Occupancy-based Policy Gradient: Estimation, Convergence, and Optimality (NeurIPS 2024 · similarity 0.86)
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.