Research map / Reinforcement learning / Visual and representation RL
Learning transferable motor skills with hierarchical latent mixture policies
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 hierarchical skills for zero-shot transfer instead of low-level action control | HeLMS · Sparse staged stacking, object set 4 | Staged reward: 0.004 -> 0.05 (BC to HeLMS) | Helped |
| Use hierarchical skills for zero-shot transfer instead of low-level action control | HeLMS · Zero-shot exploration, object set 4 | Grasp state coverage: 0.004 -> 1.10 (MPO to HeLMS); 0.0 -> 1.10 (NPMP to HeLMS) | Helped |
| Use HeLMS instead of MPO from scratch | Sparse fully-rewarded stacking tasks · Fully sparse stacking reward | Reward | Helped |
| Use HeLMS for transfer to unseen objects | Red-on-blue stacking · Object sets 1, 2, 3, and 5 | Asymptotic performance | Mixed |
| Use HeLMS skills for a new compositional task | Pyramid task · Pyramid stacking | Asymptotic performance | Helped |
| Transfer state-based skills to a vision-based policy with skill regularisation | HeLMS-mix · Vision-based stacking | Sample efficiency | Helped |
| Add discrete skill choices while keeping continuous latent variation (ablation) | HeLMS · Sparse staged stacking, easy and hard object sets | Sample efficiency | Helped |
| Add continuous latent variation to discrete skills (ablation) | HeLMS · Sparse staged stacking, hard object set 3 | Performance | Helped |
| Use offline KL regularisation of mid-level components to N(0, I) (ablation) | HeLMS-mix transfer · Hard transfer to object set 3 | Performance | Mixed |
| Remove online mid-level KL regularisation (ablation) | HeLMS-mix with offline KL regularisation · Hard transfer to object set 3 | Performance | Hurt |
| Remove offline mid-level KL regularisation (ablation) | HeLMS-mix · Easy and hard transfer | Sensitivity to online KL regularisation | No clear effect |
| Replace NPMP's AR(1) prior with N(0, I) and add RL KL regularisation (ablation) | NPMP · Manipulation across five object sets | Performance | Helped |
Closest papers by meaning
- QueST: Self-Supervised Skill Abstractions for Learning Continuous Control (NeurIPS 2024 · similarity 0.90)
- Latent Policy Barrier: Learning Robust Visuomotor Policies by Staying In-Distribution (NeurIPS 2025 · similarity 0.89)
- Rethinking Latent Redundancy in Behavior Cloning: An Information Bottleneck Approach for Robot Manipulation (ICML 2025 · similarity 0.88)
- Robust Policy Learning via Offline Skill Diffusion (AAAI 2024 · similarity 0.88)
- Learning Options for Compositional Motor Control with Adapter Banks (NeurIPS 2026 · similarity 0.88)
- Hierarchical Equivariant Policy via Frame Transfer (ICML 2025 · 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.