Research map / Reinforcement learning / Visual and representation RL
Phase-Parametric Policies for Reinforcement Learning in Cyclic Environments
Arjun Sharma, Kris Kitani
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 |
|---|---|---|---|
| add phase-parametric layers instead of phase as input | DRQN · Flying with Wind pwind=1.0 | return: -13.7 -> -2.9 | Improved |
| add phase-parametric layers instead of phase as input | DRQN · Flying with Wind pwind=0.0 | return: 2.0 -> -1.6 | Hurt |
| add phase-parametric layers instead of phase as input | DQN · Flying with Wind pwind=0.6 | return: 0.4 -> 0.5 | No clear effect |
| add phase-parametric layers instead of phase as input | DQN · Flying with Wind pwind=0.0 | return: 2.4 -> -0.5 | Hurt |
| add phase-parametric layers instead of phase as input | DDPG · Hopper short training | average return: 2133.34 -> 1397.72 | Hurt |
| train longer (40000 vs 10000 episodes) | Phase-DDPG · Hopper | average return: 1397.72 -> 2630.16 | Improved |
| train longer (40000 vs 10000 episodes) | DDPG-IP · Hopper | average return: 2133.34 -> 2559.47 | Improved |
| train longer (40000 vs 10000 episodes) | Phase-DDPG · Walker | average return: 2532.24 -> 4491.60 | Improved |
| initialize control weights from a DDPG-NP model then fine-tune (ablation) | Phase-DDPG · Hopper | average return: 1397.72 -> 3325.57 | Improved |
| initialize control weights from a DDPG-NP model then fine-tune (ablation) | Phase-DDPG · Walker | average return: 2532.24 -> 5124.44 | Improved |
| use feed-forward Phase-DQN instead of recurrent Phase-DRQN | Phase-Parametric Q-network · Freeway On-Ramp R4:8,0:0 | return: 0 -> 8 | Improved |
| vary wind probability pwind from 0.0 to 1.0 | Phase-DRQN · Flying with Wind | return: -1.6 -> -2.9 | No clear effect |
Closest papers by meaning
- From Ticks to Flows: Dynamics of Neural Reinforcement Learning in Continuous Environments (ICLR 2026 · similarity 0.86)
- Reparameterized Policy Learning for Multimodal Trajectory Optimization (ICML 2023 · similarity 0.86)
- Policy Gradient Methods in the Presence of Symmetries and State Abstractions (JMLR 2024 · similarity 0.86)
- Abstract and Explore: A Novel Behavioral Metric with Cyclic Dynamics in Reinforcement Learning (AAAI 2024 · similarity 0.85)
- The Ladder in Chaos: Improving Policy Learning by Harnessing the Parameter Evolving Path in A Low-dimensional Space (NeurIPS 2024 · similarity 0.85)
- The Horizon Threshold in Cooperative Multi-Agent Reward-Free Exploration (NeurIPS 2026 · similarity 0.85)
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.
