Research map / Reinforcement learning / Policy gradient and actor-critic
Learning Guidance Rewards with Trajectory-space Smoothing
Tanmay Gangwani, Yuan Zhou, Jian Peng
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 |
|---|---|---|---|
| replace environmental rewards with IRCR guidance rewards | SAC · Peg insertion | performance: $104 \scriptstyle{\pm 4}$ -> $160 \scriptstyle{\pm 7}$ | Improved |
| replace environmental rewards with IRCR guidance rewards | TD3 · Hopper | performance: $235 \scriptstyle{\pm 37}$ -> $2981 \scriptstyle{\pm 114}$ | Improved |
| replace environmental rewards with IRCR guidance rewards | TD3 · Half-Cheetah | performance: $-123 \scriptstyle{\pm 50}$ -> $6844 \scriptstyle{\pm 918}$ | Improved |
| use n-step returns (n=10) instead of one-step returns | SAC with environmental rewards · Hopper and Half-Cheetah | performance: $626 \scriptstyle{\pm 281}$ on Hopper; $-42 \scriptstyle{\pm 29}$ on Half-Cheetah | No clear effect |
| use λ-returns (λ=0.9) instead of one-step returns | SAC with environmental rewards · Hopper and Half-Cheetah | performance: $566 \scriptstyle{\pm 101}$ on Hopper; $-112 \scriptstyle{\pm 206}$ on Half-Cheetah | No clear effect |
| replace environmental rewards with guidance rewards | SAC · MuJoCo locomotion tasks with episodic rewards | sample efficiency | Improved |
| replace environmental rewards with guidance rewards | MA-TD3 · Rover Domain, coupling factors 1–4 | percentage of POIs harvested | Improved |
| use IRCR guidance rewards with distributional C51 | MA-C51 · Rover Domain | percentage of POIs harvested | Improved |
Closest papers by meaning
- Learning Process Rewards via Visitation Matching for Efficient RL (NeurIPS 2026 · similarity 0.89)
- Learning Long-Term Reward Redistribution via Randomized Return Decomposition (ICLR 2022 · similarity 0.88)
- DISCOVER: Automated Curricula for Sparse-Reward Reinforcement Learning (NeurIPS 2025 · similarity 0.87)
- STAS: Spatial-Temporal Return Decomposition for Solving Sparse Rewards Problems in Multi-agent Reinforcement Learning (AAAI 2024 · similarity 0.87)
- Sequence Compression Speeds Up Credit Assignment in Reinforcement Learning (ICML 2024 · similarity 0.86)
- Long-Horizon Q-Learning: Accurate Value Learning via n-Step Inequalities (NeurIPS 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.
