Research map / Reinforcement learning / Visual and representation RL
Accelerating Robotic Reinforcement Learning via Parameterized Action Primitives
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 |
|---|---|---|---|
| Replace raw actions with parameterized robot action primitives | Dreamer on Kitchen · Kettle | final success rate: 0.8 -> .93 | Helped |
| Replace raw actions with parameterized robot action primitives | Dreamer on Kitchen · Hinge Cabinet | final success rate: 0.0 -> 1.0 | Helped |
| Replace raw actions with parameterized robot action primitives | SAC on Kitchen · Light Switch | final success rate: .86 -> .67 | Hurt |
| Replace raw actions with parameterized robot action primitives | PPO on Kitchen · Kettle | final success rate: .33 -> 1.0 | Helped |
| Replace raw actions with parameterized robot action primitives | RAPS on Kitchen · performance across tasks and RL algorithms | final success rate | Mixed |
| Add the dummy primitive to the hand-designed primitive library (ablation) | RAPS on Kitchen · Kitchen tasks | performance | Helped |
| Use a 6DOF rather than 3DOF dummy primitive (ablation) | RAPS on Kitchen · Kitchen tasks | performance | No clear effect |
| Use RAPS instead of raw actions for real-robot trajectories | Randomly initialized RL agent on xArm 6 · xArm 6 trajectories | execution speedup: 32x | Helped |
| Transfer the trained RAPS policy from xARM 7 to xARM 6 without further training | RAPS door-opening policy · Robosuite door opening | success rate: 100% | Helped |
| Explore with RAPS primitives before task-specific fine-tuning instead of raw actions | Plan2explore followed by Dreamer · Kitchen hinge cabinet | fine-tuning time to solve: 1 hour | Helped |
| Use RAPS rather than raw actions for unsupervised exploration and subsequent fine-tuning | Plan2explore followed by Dreamer · Kitchen top-left burner and hinge cabinet | fine-tuning performance | Helped |
| Use RAPS rather than Raw Actions | Dreamer on Metaworld · 50 Metaworld tasks | tasks solved: 21 -> 43 | Helped |
Closest papers by meaning
- Deep Reinforcement Learning for Robotics: A Survey of Real-World Successes (AAAI 2025 · similarity 0.90)
- Evolutionary Reinforcement Learning with Parameterized Action Primitives for Diverse Manipulation Tasks (AAAI 2025 · similarity 0.88)
- PEAR: Primitive Enabled Adaptive Relabeling for Boosting Hierarchical Reinforcement Learning (ICLR 2025 · similarity 0.88)
- Dual RL: Unification and New Methods for Reinforcement and Imitation Learning (ICLR 2024 · similarity 0.88)
- Plan-Seq-Learn: Language Model Guided RL for Solving Long Horizon Robotics Tasks (ICLR 2024 · similarity 0.88)
- NS-VLA: Towards Neuro-Symbolic Vision-Language-Action Models (NeurIPS 2026 · similarity 0.88)
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.