Research map / Reinforcement learning / Visual and representation RL
Human-AI Shared Control via Policy Dissection
Quanyi Li, Zhenghao Peng, Haibin Wu, Lan Feng, Bolei Zhou
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 human activation of motor primitives at takeover start | PPO policy · MetaDrive unseen test scenes | success rate: 0.75 -> 0.95 | Improved |
| Add human activation of motor primitives at takeover start | PPO policy · MetaDrive unseen test scenes | episodic cost: 3.37 -> 0.05 | Improved |
| Add human activation of motor primitives at takeover start | SAC policy · MetaDrive unseen test scenes | success rate: 0.75 -> 0.95 | Improved |
| Add human activation of motor primitives at takeover start | SAC policy · MetaDrive unseen test scenes | episodic cost: 5.75 -> 0.80 | Improved |
| Add human activation of motor primitives | Proprioceptive-state A1 policy · Pybullet-A1 obstacle avoidance | moving distance: 6.55 -> 16.94 | Improved |
| Use human activation of motor primitives rather than a vision-trained policy | Pybullet-A1 obstacle avoidance · Pybullet-A1 obstacle avoidance | moving distance: 16.94; 18.64 | Mixed |
| Use ReLU instead of tanh activation (ablation) | PPO with Policy Dissection · MetaDrive unseen test scenes | success rate: 0.95 -> 0.88 | Hurt |
| Use 6 hidden layers instead of 2 (ablation) | PPO with Policy Dissection · MetaDrive unseen test scenes | success rate: 0.95 -> 0.9 | Hurt |
| Replace direct yaw goal input with activation of the yaw motor primitive | ANYmal-C goal-conditioned policy · ANYmal-C yaw tracking | heading tracking precision | No clear effect |
Closest papers by meaning
- Beyond Single Stationary Policies: Meta-Task Players as Naturally Superior Collaborators (NeurIPS 2024 · similarity 0.88)
- Shared Autonomy with IDA: Interventional Diffusion Assistance (NeurIPS 2024 · similarity 0.86)
- Discovering Creative Behaviors through DUPLEX: Diverse Universal Features for Policy Exploration (NeurIPS 2024 · similarity 0.85)
- Efficient Language-instructed Skill Acquisition via Reward-Policy Co-Evolution (AAAI 2025 · similarity 0.85)
- DUO: Diverse, Uncertain, On-Policy Query Generation and Selection for Reinforcement Learning from Human Feedback (AAAI 2025 · similarity 0.85)
- Latent Policy Barrier: Learning Robust Visuomotor Policies by Staying In-Distribution (NeurIPS 2025 · 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.
