Research map / Reinforcement learning / Visual and representation RL
Variational Empowerment as Representation Learning for Goal-Conditioned Reinforcement Learning
Jongwook Choi, Archit Sharma, Honglak Lee, Sergey Levine, Shixiang Shane Gu
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 Posterior HER to DIAYN (ablation) | DIAYN · MuJoCo locomotion; large discrete goal spaces | LGR(z) accuracy: 0.801 -> 0.844 (HalfCheetah, |G|=200); 0.113 -> 0.263 (Ant, |G|=200); 0.337 -> 0.766 (Humanoid, |G|=200) | Improved |
| Add Posterior HER to DIAYN (ablation) | DIAYN · MuJoCo locomotion, |G|=10–1000 | LGR(z) accuracy: HalfCheetah, |G|=10: 0.963 -> 0.934; Ant, |G|=10: 0.806 -> 0.889 | Mixed |
| Add spectral normalization to the discriminator (ablation) | DIAYN · 2D PointMass, |G|=10 and |G|=20 | LGR(z): 0.94 -> 0.96; 0.91 -> 0.96 | Improved |
| Add spectral normalization to the discriminator (ablation) | DIAYN · 2D PointMass, |G|=50 | LGR(z): 0.83 -> 0.78 | Hurt |
| Learn a global adaptive covariance instead of fixing the goal-reward variance (ablation) | GCRL · Windy PointMass | goal-reaching performance in controllable dimension | Improved |
| Use a linear learned mapping from observations to latent goals (ablation) | GCRL · 2D PointMass with random linear observation projection | recovery of intrinsic dimensions | Improved |
| Add spectral normalization to a Gaussian discriminator (ablation) | VGCRL-Gaussian · MuJoCo locomotion | LGR(z) | Improved |
Closest papers by meaning
- Unifying Goal-Conditioned RL and Unsupervised Skill Learning via Control-Maximization (NeurIPS 2026 · similarity 0.90)
- Discrete Compositional Representations as an Abstraction for Goal Conditioned Reinforcement Learning (NeurIPS 2022 · similarity 0.90)
- Offline Goal-conditioned Reinforcement Learning with Quasimetric Representations (NeurIPS 2025 · similarity 0.89)
- Improving the Continuity of Goal-Achievement Ability via Policy Self-Regularization for Goal-Conditioned Reinforcement Learning (ICML 2025 · similarity 0.88)
- Skill-aware Mutual Information Optimisation for Zero-shot Generalisation in Reinforcement Learning (NeurIPS 2024 · similarity 0.87)
- Towards an Information Theoretic Framework of Context-Based Offline Meta-Reinforcement Learning (NeurIPS 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.
