Research map / Reinforcement learning / Policy gradient and actor-critic
Model-Free Trajectory-based Policy Optimization with Monotonic Improvement
Riad Akrour, Abbas Abdolmaleki, Hany Abdulsamad, Jan Peters, Gerhard Neumann
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 |
|---|---|---|---|
| use MOTO with entropy reduction constant beta0 = .1 instead of .5 (ablation) | MOTO · Double link swing-up | return | Improved |
| replace GPS (linearized dynamics) with MOTO (quadratic Q-function) | trajectory optimization · Double link swing-up | return | Improved |
| replace GPS with MOTO on a task with joint limits | trajectory optimization · Quad link swing-up (joint limits) | return | Improved |
| add the entropy constraint (ablation) | MOTO · Double link swing-up (M = 20) | return | Mixed |
| add importance sampling (ablation) | MOTO · Double link swing-up (M = 20) | convergence speed | Improved |
| use dynamic programming Q-estimate instead of Monte-Carlo estimate (ablation) | MOTO · Double link swing-up (M = 20) | return | Improved |
| reduce rollouts per iteration M from 20 to 5 (ablation) | MOTO · Double link swing-up | sample efficiency | Improved |
| replace DMP-based REPS with MOTO | robot table tennis (no noise) · Robot table tennis | convergence speed | Improved |
| replace DMP-based REPS/MORE with MOTO | robot table tennis · Robot table tennis (final policy quality) | return | Hurt |
| replace DMP-based methods with MOTO under bounce noise | robot table tennis · Robot table tennis (noisy bounce) | return | Improved |
| replace default neural-network TRPO with MOTO | multi-link swing-up · Double and quad link swing-up | return | Improved |
| give TRPO a quadratic V-function (ablation) | TRPO · Double link swing-up | early return | Mixed |
Closest papers by meaning
- Model-Free Trajectory Optimization for Reinforcement Learning (ICML 2016 · similarity 0.94)
- Truly Deterministic Policy Optimization (NeurIPS 2022 · similarity 0.89)
- Actor-Free Continuous Control via Structurally Maximizable Q-Functions (NeurIPS 2025 · similarity 0.87)
- Deterministic Policy Gradient Primal-Dual Methods for Continuous-Space Constrained MDPs (AAAI 2025 · similarity 0.86)
- Last-Iterate Global Convergence of Policy Gradients for Constrained Reinforcement Learning (NeurIPS 2024 · similarity 0.86)
- Absolute State-wise Constrained Policy Optimization: High-Probability State-wise Constraints Satisfaction (NeurIPS 2026 · 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.
