Research map / Reinforcement learning / Visual and representation RL
Autonomous Reinforcement Learning via Subgoal Curricula
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 VaPRL starting-state curriculum instead of naive RL | SAC-based persistent RL · table-top rearrangement, sawyer door closing, hand manipulation | success rate / sample efficiency | Helped |
| use VaPRL curriculum instead of forward-backward controller FBRL | persistent RL · three continuous control environments | sample efficiency | Helped |
| use VaPRL instead of perturbation controller R3L | persistent RL · three continuous control environments | sample efficiency | Helped |
| VaPRL without episodic resets versus oracle RL with resets | VaPRL · sawyer door closing | number of environment resets: 500x fewer resets | Helped |
| VaPRL without episodic resets versus oracle RL with resets | VaPRL · table-top rearrangement and hand manipulation | number of environment resets: 1000x fewer resets | Helped |
| VaPRL (reset-free) compared to oracle RL on sparse-reward tasks | VaPRL · sawyer door closing | success rate | No clear effect |
| VaPRL (reset-free) compared to oracle RL with resets | VaPRL · table-top rearrangement | success rate | Helped |
| VaPRL compared to oracle RL on dense-reward hand manipulation | VaPRL · hand manipulation | success rate | Mixed |
| give VaPRL direct resets to the curriculum state C(g) (VaPRL + reset) (ablation) | VaPRL · table-top rearrangement | success rate | Helped |
| give FBRL resets to initial distribution (oracle RL / FBRL + reset) and R3L uniform resets (ablation) | FBRL and R3L · table-top rearrangement | success rate | Hurt |
| run the whole training with VaPRL: total human interventions | VaPRL · all environments | total interventions: less than 20 total interventions | Helped |
Closest papers by meaning
- Autonomous Reinforcement Learning: Formalism and Benchmarking (ICLR 2022 · similarity 0.90)
- Demonstration-free Autonomous Reinforcement Learning via Implicit and Bidirectional Curriculum (ICML 2023 · similarity 0.90)
- Oracle-Efficient Reinforcement Learning for Max Value Ensembles (NeurIPS 2024 · similarity 0.87)
- Autonomous Option Invention for Continual Hierarchical Reinforcement Learning and Planning (AAAI 2025 · similarity 0.87)
- Fast TRAC: A Parameter-Free Optimizer for Lifelong Reinforcement Learning (NeurIPS 2024 · similarity 0.87)
- Hybrid Inverse Reinforcement Learning (ICML 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.