Research map / Reinforcement learning / Visual and representation RL
Reset-Free Lifelong Learning with Skill-Space Planning
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 LiSP rather than MPC-Long for offline planning | Lifelong Hopper · Forward Hop -- Fast | performance: 0.27 $\pm$ 0.10 -> 0.84 $\pm$ 0.02 | Helped |
| use LiSP rather than MPC-Long for offline planning | Lifelong Hopper · Forward Hop -- Slow | performance: 0.49 $\pm$ 0.42 -> 0.88 $\pm$ 0.05 | Helped |
| use LiSP rather than MPC-Long for offline planning | Lifelong Hopper · Backward Hop -- Slow | performance: 0.38 $\pm$ 0.16 -> 0.81 $\pm$ 0.01 | Helped |
| use the first 200000 timesteps instead of the 1m dataset (ablation) | LiSP · Lifelong Hopper, Forward Hop -- Fast | performance: 0.84 $\pm$ 0.02 -> 0.80 $\pm$ 0.01 | Hurt |
| use the first 200000 timesteps instead of the 1m dataset (ablation) | LiSP · Lifelong Hopper, Forward Hop -- Slow | performance: 0.88 $\pm$ 0.05 -> 0.80 $\pm$ 0.01 | Hurt |
| use the first 200000 timesteps instead of the 1m dataset (ablation) | LiSP · Lifelong Hopper, Backward Hop -- Slow | performance: 0.81 $\pm$ 0.01 -> 0.79 $\pm$ 0.01 | Hurt |
| plan with skills rather than actions (ablation) | LiSP · Lifelong Hopper | planning stability | Helped |
| use a skill-practice distribution to select skills to practice (ablation) | LiSP · 2D Minecraft | asymptotic performance | Helped |
| practice skills that maximize intrinsic reward rather than choosing skills naively (ablation) | LiSP · 2D Minecraft | learning performance | Helped |
| resample skills every K timesteps rather than every timestep (ablation) | DADS-Off · Lifelong Ant | skill diversity | Hurt |
| use model-generated skill rollouts and the skill-practice distribution rather than DADS (ablation) | LiSP · episodic Hopper | average height (empowerment proxy): ≈ 5× sample efficiency increase | Helped |
Closest papers by meaning
- Lifelong Inverse Reinforcement Learning (NeurIPS 2018 · similarity 0.89)
- Fast TRAC: A Parameter-Free Optimizer for Lifelong Reinforcement Learning (NeurIPS 2024 · similarity 0.87)
- Continual Reinforcement Learning by Planning with Online World Models (ICML 2025 · similarity 0.86)
- Skill-SD: Skill-Conditioned Self-Distillation for Multi-turn LLM Agents (NeurIPS 2026 · similarity 0.86)
- When Do Skills Help Reinforcement Learning? A Theoretical Analysis of Temporal Abstractions (ICML 2024 · similarity 0.86)
- Skill-Pro: Learning Reusable Skills from Experience via Non-Parametric PPO for LLM Agents (ICML 2026 · 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.