Research map / Reinforcement learning / Policy gradient and actor-critic
A Long $N$-step Surrogate Stage Reward for Deep Reinforcement Learning
Junmin Zhong, Ruofan Wu, Jennie Si
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 LNSS surrogate reward (N=100) | DDPG · DMC Humanoid Walk, no noise | average reward: 1.3 -> 250.9 | Improved |
| add LNSS surrogate reward (N=100) | TD3 · DMC Acrobot SwingUp, no noise | average reward: 5.2 -> 114.8 | Improved |
| add LNSS surrogate reward (N=100) | D4PG · DMC Acrobot SwingUp, no noise | average reward: 133.5 -> 406.3 | Improved |
| add LNSS surrogate reward (N=100) | DDPG · DMC Finger TurnHard, no noise | average reward: 301.8 -> 943.3 | Improved |
| add LNSS surrogate reward (N=100) | TD3 · DMC Fish Swim, no noise | average reward: 483.1 -> 676.8 | Improved |
| replace 5-step n-step return with LNSS | TD3 · DMC Fish Swim, no noise | average reward: 731.7 -> 676.8 | Hurt |
| use 5-step n-step return (n5) instead of single step | TD3 · DMC Humanoid Walk, no noise | average reward: 184.2 -> 1.01 | Hurt |
| use 5-step n-step return (n5) instead of single step | DDPG · DMC Fish Swim, no noise | average reward: 272.9 -> 150.5 | Hurt |
| add LNSS surrogate reward under 10% reward noise | DDPG · DMC Humanoid Walk, 10% noise | average reward: 0.76 -> 235.4 | Improved |
| add LNSS surrogate reward under 10% reward noise | TD3 · DMC Humanoid Walk, 10% noise | average reward: 1.5 -> 200.2 | Improved |
| increase LNSS horizon N from 5, 10 to 50, 100 (ablation) | TD3-LNSS · DMC tasks | final reward and learning speed | Improved |
| increase LNSS N to 100 (vs N = 5, 50) (ablation) | TD3-LNSS · DMC tasks | cv(Q) | Improved |
Closest papers by meaning
- Robustness Verification of Deep Reinforcement Learning Based Control Systems Using Reward Martingales (AAAI 2024 · similarity 0.86)
- Progress Reward Model for Reinforcement Learning via Large Language Models (NeurIPS 2025 · similarity 0.86)
- Long-Horizon Q-Learning: Accurate Value Learning via n-Step Inequalities (NeurIPS 2026 · similarity 0.85)
- Random Latent Exploration for Deep Reinforcement Learning (ICML 2024 · similarity 0.85)
- MaxInfoRL: Boosting exploration in reinforcement learning through information gain maximization (ICLR 2025 · similarity 0.85)
- Enhancing Robustness in Deep Reinforcement Learning: A Lyapunov Exponent Approach (NeurIPS 2024 · similarity 0.85)
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.
