Research map / Reinforcement learning / Visual and representation RL
Keeping Your Distance: Solving Sparse Reward Tasks Using Self-Balancing Shaped Rewards
Alexander Trott, Stephan Zheng, Caiming Xiong, Richard Socher
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 Sibling Rivalry self-balancing reward to PPO with distance-to-goal shaping | PPO · 2D point maze | goal completion rate | Improved |
| Add Sibling Rivalry to PPO high-level controller | PPO · U-shaped Ant Maze | goal completion rate | Improved |
| Add Sibling Rivalry to PPO | PPO · 13x13 pixel-grid bitmap | goal completion rate | Improved |
| Add Sibling Rivalry to PPO | PPO · Minecraft 3D construction | exact-match success and structure completion | Improved |
| Use ICM curiosity on sparse reward instead of SR | PPO+ICM · Ant Maze | runs discovering goal: 2 in 5 | Hurt |
| Use HER instead of SR | DDPG+HER · Ant Maze | runs discovering goal: 3 of the 5 | Hurt |
| Use HER on discrete pixel-grid | HER · 13x13 pixel-grid | learning speed | Improved |
| Use ICM on discrete pixel-grid | PPO+ICM · 13x13 pixel-grid | goal completion | Hurt |
| Vary inclusion threshold epsilon from 0 to 10 (too low) (ablation) | Sibling Rivalry · 2D point maze | success rate | Hurt |
| Vary epsilon (too high) (ablation) | Sibling Rivalry · 2D point maze | success rate | Hurt |
| Use independently sampled start states for sibling rollouts (ablation) | Sibling Rivalry · 2D point maze | convergence speed | Improved |
| Independent start states for siblings (ablation) | Sibling Rivalry · Pixel-grid | goal completion | No clear effect |
Closest papers by meaning
- DISCOVER: Automated Curricula for Sparse-Reward Reinforcement Learning (NeurIPS 2025 · similarity 0.90)
- Action-Dependent Optimality-Preserving Reward Shaping (ICML 2025 · similarity 0.89)
- Highly Efficient Self-Adaptive Reward Shaping for Reinforcement Learning (ICLR 2025 · similarity 0.88)
- Catching Two Birds with One Stone: Reward Shaping with Dual Random Networks for Balancing Exploration and Exploitation (ICML 2025 · similarity 0.88)
- Auto MC-Reward: Automated Dense Reward Design with Large Language Models for Minecraft (CVPR 2024 · similarity 0.87)
- Reward Shaping for Reinforcement Learning with An Assistant Reward Agent (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.
