Research map / Reinforcement learning / Visual and representation RL
Continuously Discovering Novel Strategies via Reward-Switching Policy Optimization
Zihan Zhou, Wei Fu, Bingliang Zhang, Yi Wu
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 |
|---|---|---|---|
| Use trajectory filtering and switch to intrinsic rewards on rejected trajectories (ablation) | RSPO · 4-Goals Hard | highest expected return | Improved |
| Replace trajectory filtering with the soft no-switch objective (ablation) | RSPO · 4-Goals Hard | highest expected return | Hurt |
| Remove intrinsic rewards (ablation) | RSPO · Monster-Hunt | discovery of distinct Nash equilibria | Hurt |
| Remove reward switching (ablation) | RSPO · Monster-Hunt | discovery of distinct Nash equilibria | Hurt |
| Use behavior-driven intrinsic reward only, without reward-driven intrinsic reward (ablation) | RSPO · Monster-Hunt | discovery of the optimal Nash equilibrium | Hurt |
| Add reward-driven intrinsic reward to behavior-driven exploration (ablation) | RSPO · Monster-Hunt | discovery of the optimal Nash equilibrium | Improved |
| Add smoothed switching (ablation) | RSPO · Escalation | training stability | Improved |
| Use iterative optimization rather than population-based training with the same intrinsic reward | RSPO · MuJoCo | policy diversity | Improved |
| Set the novelty threshold too low (ablation) | RSPO · Humanoid | Population Diversity: 0.5: 0.604; 1.0: 0.676; 1.5: 0.687; 2.0: 0.748 | Hurt |
| Increase the novelty threshold α from 0.5 to 2.0 (ablation) | RSPO · Humanoid | Population Diversity: 0.5: 0.604; 1.0: 0.676; 1.5: 0.687; 2.0: 0.748 | Improved |
| Vary behavior-driven intrinsic-reward scale (ablation) | RSPO · Humanoid | Population Diversity: 0.5: 0.640; 1.0: 0.621; 5.0: 0.697; 10.0: 0.697 | Mixed |
| Turn off intrinsic rewards (ablation) | RSPO · 2m_vs_1z | convergence | Hurt |
Closest papers by meaning
- Decoupling Exploration and Policy Optimization: Uncertainty Guided Tree Search for Hard Exploration (NeurIPS 2026 · similarity 0.88)
- Action-Dependent Optimality-Preserving Reward Shaping (ICML 2025 · similarity 0.87)
- DGPO: Discovering Multiple Strategies with Diversity-Guided Policy Optimization (AAAI 2024 · similarity 0.87)
- R*: Efficient Reward Design via Reward Structure Evolution and Parameter Alignment Optimization with Large Language Models (ICML 2025 · similarity 0.87)
- DISCOVER: Automated Curricula for Sparse-Reward Reinforcement Learning (NeurIPS 2025 · similarity 0.86)
- First-Explore, then Exploit: Meta-Learning to Solve Hard Exploration-Exploitation Trade-Offs (NeurIPS 2024 · 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.
