Research map / Reinforcement learning / Visual and representation RL
Continuous Action Reinforcement Learning From a Mixture of Interpretable Experts
Riad Akrour, Davide Tateo, Jan Peters
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 number of clusters K | Metric · 7 continuous action tasks | return: 5 -> 20 clusters approaches neural network policy | Improved |
| use Metric mixture policy instead of linear policy | TRPO linear policy · 7 continuous action tasks | return: as little as 5 clusters | Improved |
| use Metric instead of TRPO/PPO neural network, high-dimensional pybullet tasks | TRPO/PPO · pybullet locomotion tasks | return: matches TRPO and PPO with fewer parameters | No clear effect |
| use small K for interpretability | Metric · 7 continuous action tasks | return: outperformed by both deep RL baselines on all tasks | Hurt |
| increase K to 40 on AntBulletEnv-v0 versus baselines | Metric · AntBulletEnv-v0 | return: 378, 748 and 1488 parameters vs 6544 | Improved |
| select prototypical states from trajectory data instead of gradient-learned centers (ablation) | Metric · pybullet locomotion tasks | gait learning: none learned a gait vs walking gaits with 10 clusters | Improved |
| add entropy constraint to differentiable prototype PPO/TRPO (ablation) | Metric · pybullet locomotion tasks | return: no gait learned -> competent policy after prototypes found | Improved |
| use trajectory-selected prototypes instead of differentiable ones (ablation) | Metric · pybullet locomotion tasks | interpretability | Improved |
Closest papers by meaning
- ORCAID: Oblique Rule-Based Continuous-Action Interpretation for Deep RL Policies (NeurIPS 2026 · similarity 0.89)
- Autonomous Option Invention for Continual Hierarchical Reinforcement Learning and Planning (AAAI 2025 · similarity 0.89)
- SMoSE: Sparse Mixture of Shallow Experts for Interpretable Reinforcement Learning in Continuous Control Tasks (AAAI 2025 · similarity 0.89)
- Overcoming Slow Decision Frequencies in Continuous Control: Model-Based Sequence Reinforcement Learning for Model-Free Control (ICLR 2025 · similarity 0.88)
- Excluding the Irrelevant: Focusing Reinforcement Learning through Continuous Action Masking (NeurIPS 2024 · similarity 0.87)
- DUO: Diverse, Uncertain, On-Policy Query Generation and Selection for Reinforcement Learning from Human Feedback (AAAI 2025 · 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.
