Research map / Reinforcement learning / Visual and representation RL
Meta-learning Parameterized Skills
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 |
|---|---|---|---|
| Add trajectory-centric smoothness loss to off-policy meta-RL skill learning (ablation) | Parameterized skills for Ant-mix and Make Coffee · Long-horizon Ant-mix and Make Coffee | overall performance | Helped |
| Use HPS instead of P-DQN with the same pretrained skills | MLPS parameterized skills · Ant obstacle course | learning efficiency | Helped |
| Use HPS instead of HyAR with the same pretrained skills | MLPS parameterized skills · Ant obstacle course and Make Coffee | overall performance | Helped |
| Add trajectory-centric smoothness loss (ablation) | Low-level meta-policy · Ant tasks | meta-test performance | Helped |
| Add trajectory-centric smoothness loss (ablation) | Low-level meta-policy · Coffee tasks | meta-test performance | No clear effect |
| Increase smoothness loss of the learned skill embeddings (ablation) | MLPS coffee-push and coffee-pull skills · Long-horizon coffee task | overall performance | Hurt |
| Use sequential trajectory encoder and compute DTW smoothness on complete trajectories (ablation) | MLPS · Ant-goal | learning speed | Hurt |
| Use sequential trajectory encoder and compute DTW smoothness on complete trajectories (ablation) | MLPS · Ant-goal | final performance | No clear effect |
| Replace the joint critic with separate critics for discrete and continuous actions (ablation) | HPS · Coffee-make long-horizon task | learning speed | Hurt |
| Use epsilon-greedy exploration instead of Gumbel-softmax exploration (ablation) | HPS · Ant obstacle course 10b-3c | performance | Hurt |
| Use the similarity score function instead of Euclidean distance between latent embeddings (ablation) | MLPS smoothness loss · Ant-goal | performance | Mixed |
| Use dynamic time warping instead of pointwise Euclidean distance to compare trajectories (ablation) | Trajectory distance for skill learning · Ant-goal, Ant-bridge, and Ant-gather | coefficient of variation | Helped |
Closest papers by meaning
- Decoupling Meta-Reinforcement Learning with Gaussian Task Contexts and Skills (AAAI 2024 · similarity 0.88)
- Learning Parameterized Skills from Demonstrations (NeurIPS 2025 · similarity 0.88)
- Model-based Reinforcement Learning for Parameterized Action Spaces (ICML 2024 · similarity 0.87)
- Goal-Oriented Skill Abstraction for Offline Multi-Task Reinforcement Learning (ICML 2025 · similarity 0.87)
- Language-guided Skill Learning with Temporal Variational Inference (ICML 2024 · similarity 0.87)
- SkillDiffuser: Interpretable Hierarchical Planning via Skill Abstractions in Diffusion-Based Task Execution (CVPR 2024 · 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.