Research map / Reinforcement learning / Visual and representation RL
Dual Goal Representations
Seohong Park, Deepinder Mann, Sergey Levine
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 |
|---|---|---|---|
| replace original goal with dual goal representation | GCIVL · OGBench state-based (13 tasks avg) | success rate: 34 -> 41 | Improved |
| replace original goal with dual goal representation | CRL · OGBench state-based (13 tasks avg) | success rate: 33 -> 39 | Improved |
| replace original goal with dual goal representation | GCFBC · OGBench state-based (13 tasks avg) | success rate: 20 -> 23 | Improved |
| replace original goal with dual goal representation | GCIVL · cube-double-play | success rate: 35 -> 60 | Improved |
| replace original goal with dual goal representation | GCIVL · antsoccer-arena-navigate | success rate: 47 -> 31 | Hurt |
| replace original goal with dual goal representation | GCIVL · pixel-based OGBench | success rate: 66 -> 78 (visual-antmaze-medium) | Improved |
| use dual goal representation on visual puzzle tasks | GCIVL · visual-puzzle-4x4 | success rate: 65 -> 0 | Hurt |
| use early fusion of state and goal images instead of late fusion (ablation) | vanilla GCIVL · visual-puzzle-4x4 | success rate: 0 -> 65 | Improved |
| use inner product aggregation instead of symmetric metric (ablation) | Dual representation · OGBench state-based avg | success rate: 38 -> 42 | Improved |
| add separate downstream GCRL value function instead of extracting policy directly from f (ablation) | Dual representation · scene-play | success rate: 50 -> 71 | Improved |
| direct policy extraction from temporal distance function (ablation) | Dual representation · antmaze-large | success rate: 28 -> 58 | Hurt |
| use dual representation for noisy evaluation goals | GCIVL · noisy OGBench environments | success rate | Improved |
Closest papers by meaning
- Offline Goal-conditioned Reinforcement Learning with Quasimetric Representations (NeurIPS 2025 · similarity 0.91)
- Conservative Offline Goal-Conditioned Implicit V-Learning (ICML 2025 · similarity 0.89)
- Flattening Hierarchies with Policy Bootstrapping (NeurIPS 2025 · similarity 0.88)
- Reconciling Spatial and Temporal Abstractions for Goal Representation (ICLR 2024 · similarity 0.88)
- Unifying Goal-Conditioned RL and Unsupervised Skill Learning via Control-Maximization (NeurIPS 2026 · similarity 0.88)
- Goal-Conditioned On-Policy Reinforcement Learning (NeurIPS 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.
