Research map / Reinforcement learning / Visual and representation RL
A Unified Algorithm Framework for Unsupervised Discovery of Skills based on Determinantal Point Process
Jiayu Chen, Vaneet Aggarwal, Tian Lan
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 |
|---|---|---|---|
| use only the landmark-based mutual information term instead of DIAYN, VIC or VALOR objectives (ablation) | ODPP (L^IB only) · Point Room trajectories | qualitative diversity and coverage | Improved |
| add DPP coverage term L1 (ablation) | ODPP (L^IB) · Point Room trajectories | qualitative diversity and coverage | Improved |
| add DPP consistency and diversity terms L2 and L3 (ablation) | ODPP (L^IB, L1) · Point Room trajectories | qualitative diversity and coverage | Improved |
| define trajectory features with Structured DPP instead of decoder hidden layer output (ablation) | ODPP · Point Room trajectories | qualitative diversity and coverage | Improved |
| set feature dimension D=2 so ODPP mimics Laplacian-based DCO (ablation) | ODPP · Point Corridor | qualitative diversity and coverage | No clear effect |
| use normal setting D=30 with 10 options instead of D=2 (ablation) | ODPP · Point Corridor | qualitative diversity and coverage | Improved |
| use ODPP options with a learned option selector instead of PPO on raw actions | Base RL (PPO) · Point Room/Corridor goal-achieving | mean return | Improved |
| use ODPP options to initialize policy for exploration instead of baselines | VIC, DIAYN, VALOR, DCO · Ant Room/Corridor exploration | trajectory return | Improved |
| initialize option selector with the learned prior network (ablation) | option selector · downstream task | downstream return | Improved |
| increase number of simultaneously learned options to 10, 20, 40, 60 (ablation) | ODPP · Ant locomotion | trajectory reward mean and std | No clear effect |
| compare ODPP skills against DIAYN, VIC, VALOR, DADS and APS | ODPP · CartPole, AirRaid, RiverRaid | average cumulative reward | Improved |
Closest papers by meaning
- Skill Disentanglement in Reproducing Kernel Hilbert Space (AAAI 2025 · similarity 0.89)
- Explore, Discover and Learn: Unsupervised Discovery of State-Covering Skills (ICML 2020 · similarity 0.88)
- Efficient Skill Discovery via Regret-Aware Optimization (ICML 2025 · similarity 0.88)
- Constrained Ensemble Exploration for Unsupervised Skill Discovery (ICML 2024 · similarity 0.85)
- DIDI: Diffusion-Guided Diversity for Offline Behavioral Generation (ICML 2024 · similarity 0.85)
- Reward-Respecting Subtasks for Model-Based Reinforcement Learning (Abstract Reprint) (AAAI 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.
