Research map / Reinforcement learning / Visual and representation RL
CoMic: Complementary Task Learning & Mimicry for Reusable Skills
Leonard Hasenclever, Fabio Pardo, Raia Hadsell, Nicolas Heess, Josh Merel
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 |
|---|---|---|---|
| Use MLP low-level policy instead of LSTM, mixture, or product architectures | low-level policy architecture · Go To Target, Gaps, Walls transfer across data regimes | mean return on transfer tasks | Improved |
| Joint training with complementary get-up task (10% of episodes) (ablation) | MLP low-level policy on largest mocap set · Get-up and stand task | task performance | Improved |
| Joint training with ball catching task (20% of episodes) (ablation) | MLP low-level policy on largest mocap set · Ball catching task | task performance | Improved |
| Joint training with complementary tasks (ablation) | MLP low-level policy on largest mocap set · Locomotion transfer tasks | transfer performance | No clear effect |
| Stronger KL regularization on latent embedding (beta) (ablation) | MLP low-level policy, latent dimension D=20-60 · Go-to-target transfer | reward speed and asymptotic return | Mixed |
| Stronger KL regularization on latent embedding (beta=5e-4) (ablation) | MLP low-level policy · Walls transfer task | mean return | Improved |
| Latent space dimensionality D (20 to 60) (ablation) | MLP low-level policy · Tracking and transfer tasks | tracking reward and transfer return | No clear effect |
| Separate value function per reward term (ablation) | mocap tracking training · Locomotion 40min mocap tracking | learning speed | Improved |
Closest papers by meaning
- AnySkill: Learning Open-Vocabulary Physical Skill for Interactive Agents (CVPR 2024 · similarity 0.88)
- Universal Humanoid Motion Representations for Physics-Based Control (ICLR 2024 · similarity 0.87)
- Meta-Controller: Few-Shot Imitation of Unseen Embodiments and Tasks in Continuous Control (NeurIPS 2024 · similarity 0.87)
- What's the Move? Hybrid Imitation Learning via Salient Points (ICLR 2025 · similarity 0.87)
- SUGAR: A Scalable Human-Video-Driven Generalizable Humanoid Loco-Manipulation Learning Framework (NeurIPS 2026 · similarity 0.87)
- Goal-Oriented Skill Abstraction for Offline Multi-Task Reinforcement Learning (ICML 2025 · similarity 0.87)
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.
