Research map / Reinforcement learning / Visual and representation RL
Open-Ended Reinforcement Learning with Neural Reward Functions
Robert Meier, Asier Mujika
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 |
|---|---|---|---|
| remove policy feature reuse (ablation) | Ours (full) · Ant | zero-shot reward: 2506 ± 511 -> 1731 ± 634 | Hurt |
| remove value reuse (ablation) | Ours (full) · Ant | zero-shot reward: 2506 ± 511 -> 2246 ± 794 | Hurt |
| remove policy feature reuse (ablation) | Ours (full) · Humanoid | zero-shot reward: 9092 ± 1063 -> 8906 ± 616 | No clear effect |
| remove value reuse (ablation) | Ours (full) · Humanoid | zero-shot reward: 9092 ± 1063 -> 7357 ± 816 | Hurt |
| remove policy feature reuse (ablation) | Ours (full) · Humanoid | MI(s,z): 1.29 ± 0.25 -> 0.88 ± 0.19 | Hurt |
| remove value reuse (ablation) | Ours (full) · Humanoid | MI(s,z): 1.29 ± 0.25 -> 1.01 ± 0.09 | Hurt |
| remove policy feature reuse (ablation) | Ours (full) · Ant | MI(s,z): 1.33 ± 0.11 -> 1.09 ± 0.16 | Hurt |
| remove value reuse (ablation) | Ours (full) · Ant | MI(s,z): 1.33 ± 0.11 -> 1.28 ± 0.15 | No clear effect |
| remove any of policy, value or guiding transfer (ablation) | Ours (full) · 2D navigation | reward functions solved: 29.3 ± 9.5, 18.7 ± 11.6, 19 ± 8.5 | Hurt |
| Disagreement/RND intrinsic motivation instead of Ours | checkpoint skills · Ant | zero-shot reward: 2506 ± 511 -> 9 ± 155 | Hurt |
| Disagreement instead of Ours | checkpoint skills · Humanoid | zero-shot reward: 9092 ± 1063 -> 10107 ± 736 | Improved |
| higher entropy regularization (ablation) | policy · 2D navigation | state space coverage | Mixed |
Closest papers by meaning
- SkiLD: Unsupervised Skill Discovery Guided by Factor Interactions (NeurIPS 2024 · similarity 0.88)
- Efficient Language-instructed Skill Acquisition via Reward-Policy Co-Evolution (AAAI 2025 · similarity 0.88)
- Efficient Skill Discovery via Regret-Aware Optimization (ICML 2025 · similarity 0.88)
- Periodic Skill Discovery (NeurIPS 2025 · similarity 0.87)
- Skill Disentanglement in Reproducing Kernel Hilbert Space (AAAI 2025 · similarity 0.87)
- Eureka: Human-Level Reward Design via Coding Large Language Models (ICLR 2024 · similarity 0.87)
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.
