Research map / Reinforcement learning / Visual and representation RL
Active Vision Reinforcement Learning under Limited Visual Observability
Jinghuan Shang, Michael S Ryoo
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 sensorimotor reward that penalizes views with poor motor-action predictability (ablation) | SUGARL-DQN · Atari 50x50, no peripheral observation | normalized IQM: 0.421 -> 0.805 | Improved |
| Add reward balancing (ablation) | SUGARL-DQN · Atari 50x50, no peripheral observation | normalized IQM: 0.423 -> 0.805 | Improved |
| Add Persistence-of-Vision Memory (ablation) | SUGARL-DQN · Atari 50x50, no peripheral observation | normalized IQM: 0.360 -> 0.423 | Improved |
| Replace positive intrinsic reward with negative prediction-error reward (ablation) | SUGARL-DQN · Atari 50x50, no peripheral observation | normalized IQM: 0.322 -> 0.360 | Improved |
| Add joint learning of motor and sensory policies (ablation) | SUGARL-DQN · Atari 50x50, no peripheral observation | normalized IQM: 0.171 -> 0.805 | Improved |
| Use separate sensory and motor policies rather than one combined action policy | DrQ · DMC 20x20 | normalized IQM: 0.540 -> 0.686 | Improved |
| Use separate sensory and motor policies rather than one combined action policy | DrQ · DMC 30x30 | normalized IQM: 0.570 -> 0.717 | Improved |
| Use separate sensory and motor policies rather than one combined action policy | DrQ · DMC 50x50 | normalized IQM: 0.776 -> 1.052 | Improved |
| Use absolute rather than relative sensory control | SUGARL-DQN · Atari 20x20, no peripheral observation | normalized IQM: 0.367 -> 0.475 | Improved |
| Use relative rather than absolute sensory control | SUGARL-DQN · Atari 50x50, no peripheral observation | normalized IQM: 0.805 -> 0.945 | Improved |
| Train for 5M rather than 1M steps | SUGARL-DQN · Atari 50x50, no peripheral observation | normalized IQM: 0.805 -> 1.553 | Improved |
| Replace SAC entropy autotuning with fixed sensory-policy alpha | SUGARL-SAC · Atari 20x20, no peripheral observation | normalized IQM: 0.271 -> 0.424 | Improved |
Closest papers by meaning
- ACTIVE-o3 : Empowering MLLMs with Active Perception via Pure Reinforcement Learning (ICML 2026 · similarity 0.90)
- APPLE: Toward General Active Perception via Reinforcement Learning (ICLR 2026 · similarity 0.89)
- Real-World Reinforcement Learning of Active Perception Behaviors (NeurIPS 2025 · similarity 0.89)
- SOLE-R1: Video-Language Reasoning as the Sole Reward for On-Robot Reinforcement Learning (NeurIPS 2026 · similarity 0.86)
- Perception-R1: Pioneering Perception Policy with Reinforcement Learning (NeurIPS 2025 · similarity 0.86)
- Robot-R1: Reinforcement Learning for Enhanced Embodied Reasoning in Robotics (NeurIPS 2025 · 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.
