Research map / Reinforcement learning / Visual and representation RL
One Policy to Control Them All: Shared Modular Policies for Agent-Agnostic Control
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 shared modular policies with both-way message passing instead of a monolithic multi-task policy | Multi-morphology locomotion control · Walker++, Humanoid++, Cheetah++, Walker-Hopper++, Walker-Hopper-Humanoid++ | average reward | Helped |
| Use both-way message passing instead of top-down-only or bottom-up-only message passing (ablation) | Shared Modular Policies · Multi-morphology locomotion tasks | average reward | Helped |
| Use both-way message passing instead of one-way message passing (ablation) | Shared Modular Policies · Held-out morphologies | average reward | Helped |
| Remove communication between limb modules (ablation) | Shared Modular Policies · Single-agent locomotion sanity check | average reward | Hurt |
| Pass the same message to all children instead of distinct child-specific messages (ablation) | Shared Modular Policies · Multi-morphology locomotion | performance | Hurt |
| Change the graph root from the torso to the left foot (ablation) | SMP with both-way message passing · Walker2D-v2 walker variant, training at 1M timesteps | training reward: 3215.04 -> 3709.87 | Helped |
Closest papers by meaning
- Low-Rank Modular Reinforcement Learning via Muscle Synergy (NeurIPS 2022 · similarity 0.90)
- Learning to Control Self-Assembling Morphologies: A Study of Generalization via Modularity (NeurIPS 2019 · similarity 0.90)
- Make-An-Agent: A Generalizable Policy Network Generator with Behavior-Prompted Diffusion (NeurIPS 2024 · similarity 0.87)
- MeMo: Meaningful, Modular Controllers via Noise Injection (NeurIPS 2024 · similarity 0.87)
- Convergent Functions, Divergent Forms (NeurIPS 2025 · similarity 0.87)
- DittoGym: Learning to Control Soft Shape-Shifting Robots (ICLR 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.