Research map / Reinforcement learning / Visual and representation RL
HyAR: Addressing Discrete-Continuous Action Reinforcement Learning via Hybrid Action Representation
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 HyAR-TD3 instead of the strongest listed baseline | hybrid-action RL · Goal | success rate: 0.71$\pm$0.10 -> 0.78$\pm$0.03 | Helped |
| Use HyAR-TD3 instead of the strongest listed baseline | hybrid-action RL · Hard Goal | success rate: 0.44$\pm$0.05 -> 0.60$\pm$0.07 | Helped |
| Use HyAR-TD3 instead of the strongest listed baseline | hybrid-action RL · Platform | success rate: 0.94$\pm$0.10 -> 0.98$\pm$0.01 | Helped |
| Use HyAR-TD3 instead of the strongest listed baseline | hybrid-action RL · Catch Point | success rate: 0.89$\pm$0.07 -> 0.90$\pm$0.03 | Helped |
| Use HyAR-TD3 instead of the strongest listed baseline | hybrid-action RL · Hard Move (n = 8) | success rate: 0.05$\pm$0.02 -> 0.89$\pm$0.03 | Helped |
| Use element-wise product rather than concatenation (ablation) | HyAR-TD3 · Goal | success rate: 0.66$\pm$ \ 0.10 -> 0.78$\pm$ \ 0.03 | Helped |
| Use a cascaded head rather than a parallel head (ablation) | HyAR-TD3 · Goal | success rate: 0.71 $\pm$ \ 0.04 -> 0.78$\pm$ \ 0.03 | Helped |
| Add representation relabeling (ablation) | HyAR-TD3 · Goal | success rate: 0.66 $\pm$ 0.07 -> 0.78$\pm$ \ 0.03 | Helped |
| Set latent select range to 96% rather than 100% (ablation) | HyAR-TD3 · Goal | success rate: 0.62 $\pm$ 0.11 -> 0.78$\pm$ \ 0.03 | Helped |
| Set latent select range to 96% rather than 90% (ablation) | HyAR-TD3 · Goal | success rate: 0.61 $\pm$ 0.04 -> 0.78$\pm$ \ 0.03 | Helped |
| Set latent select range to 96% rather than 80% (ablation) | HyAR-TD3 · Goal | success rate: 0.08 $\pm$ 0.17 -> 0.78$\pm$ \ 0.03 | Helped |
| Use latent action dimension 6 rather than 3 (ablation) | HyAR-TD3 · Hard Move (n = 8) | success rate: 0.58$\pm$ 0.16 -> 0.89$\pm$ 0.03 | Helped |
Closest papers by meaning
- CHDP: Cooperative Hybrid Diffusion Policies for Reinforcement Learning in Parameterized Action Space (AAAI 2026 · similarity 0.89)
- Learning Action Representations for Reinforcement Learning (ICML 2019 · similarity 0.88)
- Flow Matching for Offline Reinforcement Learning with Discrete Actions (NeurIPS 2026 · similarity 0.86)
- Geometry of Neural Reinforcement Learning in Continuous State and Action Spaces (ICLR 2025 · similarity 0.86)
- Discrete Codebook World Models for Continuous Control (ICLR 2025 · similarity 0.86)
- Excluding the Irrelevant: Focusing Reinforcement Learning through Continuous Action Masking (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.