Research map / Reinforcement learning / Visual and representation RL
Learning to Shape Rewards Using a Game of Two Partners
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 ROSA reward shaping | PPO · Cartpole | learning performance | Helped |
| Add ROSA reward shaping | PPO · Gravitar | learning performance | Helped |
| Add ROSA reward shaping | PPO · Solaris | learning performance | No clear effect |
| Add ROSA reward shaping | PPO · Super Mario | learning performance | No clear effect |
| Add ROSA to vanilla PPO or TRPO (ablation) | PPO and TRPO · 25x25 Gridworld | learning performance | Helped |
| Remove the Shaper's exploration bonus (ablation) | PPO+ROSA · 25x25 Gridworld | learning performance | Helped |
| Replace RND exploration bonus with a count-based bonus (ablation) | PPO+ROSA · 25x25 Gridworld | reward-shaping effectiveness | Hurt |
| Add an RND exploration bonus to the Shaper (ablation) | PPO+ROSA · 25x25 Gridworld | reward-shaping effectiveness | Helped |
| Remove switching controls so the Shaper shapes every state (ablation) | ROSA · Maze | steps to average episode return 0.5: almost 50,000 more steps | Hurt |
| Tune the switching cost outside the effective range (ablation) | PPO+ROSA · 25x25 Gridworld | end-of-training evaluation performance: approximately (-0.1,-0.01) | Hurt |
| Adapt shaping rewards to the Controller's policy (ablation) | ROSA · Corridor Maze | alignment of shaping rewards with the optimal goal | Helped |
Closest papers by meaning
- Reward Shaping for Reinforcement Learning with An Assistant Reward Agent (ICML 2024 · similarity 0.92)
- Highly Efficient Self-Adaptive Reward Shaping for Reinforcement Learning (ICLR 2025 · similarity 0.89)
- Centralized Reward Agent for Knowledge Sharing and Transfer in Multi-Task Reinforcement Learning (NeurIPS 2025 · similarity 0.89)
- Automatic Reward Shaping from Confounded Offline Data (ICML 2025 · similarity 0.89)
- BAMDP Shaping: a Unified Framework for Intrinsic Motivation and Reward Shaping (ICLR 2025 · similarity 0.88)
- ORSO: Accelerating Reward Design via Online Reward Selection and Policy Optimization (ICLR 2025 · similarity 0.88)
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.