Research map / Reinforcement learning / Visual and representation RL
Learning by Playing Solving Sparse Reward Tasks from Scratch
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 |
|---|---|---|---|
| actively learn and schedule auxiliary intentions | SAC-X on two-block stacking · Two-block stacking | learning reliability and speed | Helped |
| learn the intention scheduler instead of scheduling uniformly (ablation) | SAC-X · Two-block stacking | learning speed and final performance | Helped |
| learn the intention scheduler instead of scheduling uniformly (ablation) | SAC-X · Banana stacking | cumulative reward | Helped |
| replace object-position features with camera pixels (ablation) | SAC-Q · Two-block stacking | learning speed | Hurt |
| replace object-position features with camera pixels (ablation) | SAC-Q · Two-block stacking | task learnability | No clear effect |
| remove intention scheduling and follow the external-reward behavior policy (ablation) | SAC-X (IUA variant) · Two-block stacking | learning speed and reliability | Hurt |
| remove auxiliary-intention scheduling and use DDPG | two-block stacking agent · Two-block stacking | task learning | Hurt |
| learn an additional external stacking task using the shared replay buffer | SAC-X · Stacking blocks both ways | episodes required for additional task: 2500 additional episodes per actor | Helped |
| use a learned scheduler instead of uniform scheduling (ablation) | SAC-X · Clean-up, all objects in box | task learning | Helped |
| train SAC-Q with auxiliary rewards and scheduled intentions on the real robot | SAC-Q · Real robot lifting | success rate: 100% | Helped |
Closest papers by meaning
- Enhanced Meta Reinforcement Learning via Demonstrations in Sparse Reward Environments (NeurIPS 2022 · similarity 0.89)
- DISCOVER: Automated Curricula for Sparse-Reward Reinforcement Learning (NeurIPS 2025 · similarity 0.88)
- Reward Shaping for Reinforcement Learning with An Assistant Reward Agent (ICML 2024 · similarity 0.88)
- Learning Process Rewards via Visitation Matching for Efficient RL (NeurIPS 2026 · similarity 0.88)
- Uncertainty-Guided Exploration and Stable Planning for Sparse-Reward Manipulation from Limited Demonstrations (ICML 2026 · similarity 0.88)
- Multi-Stage Manipulation with Demonstration-Augmented Reward, Policy, and World Model 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.