Research map / Reinforcement learning / Visual and representation RL
Improving Exploration in Evolution Strategies for Deep Reinforcement Learning via a Population of Novelty-Seeking Agents
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 |
|---|---|---|---|
| combine behavioral novelty with reward optimization (ablation) | ES on Humanoid Locomotion with Deceptive Trap · Humanoid Locomotion with Deceptive Trap | reward: 5.3 -> 31.2 | Helped |
| adapt the balance between reward and novelty during training (ablation) | ES on Humanoid Locomotion with Deceptive Trap · Humanoid Locomotion with Deceptive Trap | reward: 5.3 -> 37.1 | Helped |
| optimize novelty alone instead of reward (ablation) | ES on Humanoid Locomotion, Isotropic · Humanoid Locomotion, Isotropic | reward: 8098.5 -> 6010.5 | Hurt |
| adapt the balance between reward and novelty during training (ablation) | ES on Humanoid Locomotion, Isotropic · Humanoid Locomotion, Isotropic | reward: 8098.5 -> 7923.1 | No clear effect |
| combine novelty with reward rather than optimize novelty alone (ablation) | ES on Humanoid Locomotion, Isotropic · Humanoid Locomotion, Isotropic | reward: 6010.5 -> 6756.9 | Helped |
| combine novelty and reward using a fixed equal weighting (ablation) | ES on Atari games with premature convergence · Seaquest, Q*Bert, Freeway, Beam Rider, Frostbite | median reward | Helped |
| combine novelty and reward using a fixed equal weighting (ablation) | ES on Atari games without observed local optima · Alien, Amidar, Bank Heist, Gravitar, Montezuma's Revenge, Ms. Pacman, Zaxxon | median reward | Hurt |
| adapt the weighting between novelty and reward rather than keep it fixed (ablation) | ES on Atari · Atari games | median reward: higher median rewards than ES on 8/12 games and NSR-ES on 9/12 games | Helped |
| optimize novelty alone instead of reward (ablation) | ES on Seaquest · Seaquest | median reward: 960 -> 1044.5 | Helped |
| combine novelty with reward (ablation) | ES on Seaquest · Seaquest | median reward: 960 -> 2329.7 | Helped |
| increase the frequency of NSRA-ES weight adjustments from every 50 to every 10 generations (ablation) | NSRA-ES on Seaquest · Seaquest | median reward | Helped |
| use an adaptive novelty-reward balance instead of a fixed balance (ablation) | ES on Atari games with premature convergence · Seaquest, Q*Bert, Freeway, Beam Rider, Frostbite | median reward | Mixed |
Closest papers by meaning
- Diversity-Driven Exploration Strategy for Deep Reinforcement Learning (NeurIPS 2018 · similarity 0.92)
- Evolution-Guided Policy Gradient in Reinforcement Learning (NeurIPS 2018 · similarity 0.91)
- Effective Diversity in Population Based Reinforcement Learning (NeurIPS 2020 · similarity 0.89)
- Decoupling Exploration and Policy Optimization: Uncertainty Guided Tree Search for Hard Exploration (NeurIPS 2026 · similarity 0.88)
- Two-Stage Evolutionary Reinforcement Learning for Enhancing Exploration and Exploitation (AAAI 2024 · similarity 0.88)
- DGPO: Discovering Multiple Strategies with Diversity-Guided Policy Optimization (AAAI 2024 · 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.