Research map / Reinforcement learning / Visual and representation RL
VIP: Towards Universal Visual Reward and Representation via Value-Implicit Pre-Training
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 VIP-RWR instead of VIP-BC (ablation) | VIP with pre-trained representation · CloseDrawer | success rate (%): 50 ± { 50} -> 100 ± { 0} | Helped |
| Use VIP-RWR instead of VIP-BC (ablation) | VIP with pre-trained representation · PushBottle | success rate (%): 50 ± { 50} -> 90 ± { 30} | Helped |
| Use VIP-RWR instead of VIP-BC (ablation) | VIP with pre-trained representation · PlaceMelon | success rate (%): 10 ± { 30} -> 60 ± { 48} | Helped |
| Use VIP-RWR instead of VIP-BC (ablation) | VIP with pre-trained representation · FoldTowel | success rate (%): 20 ± { 40} -> 90 ± { 30} | Helped |
| Use VIP-RWR instead of R3M-RWR | pre-trained representation for offline RWR · CloseDrawer | success rate (%): 80 ± { 40} -> 100 ± { 0} | Helped |
| Use VIP-RWR instead of R3M-RWR | pre-trained representation for offline RWR · PushBottle | success rate (%): 70 ± { 46} -> 90 ± { 30} | Helped |
| Use VIP-RWR instead of R3M-RWR | pre-trained representation for offline RWR · PlaceMelon | success rate (%): 0 ± { 0} -> 60 ± { 48} | Helped |
| Use VIP-RWR instead of R3M-RWR | pre-trained representation for offline RWR · FoldTowel | success rate (%): 0 ± { 0} -> 90 ± { 30} | Helped |
| Train the VIP representation on task-specific data rather than pre-training on Ego4D (ablation) | VIP-RWR and VIP-BC · real-robot tasks | success rate (%): 0 ± { 0}; 0* ± { 0} | Hurt |
| Increase the MPPI optimization budget | VIP reward on FrankaKitchen · FrankaKitchen | aggregate success rate: 44% | Helped |
| Increase the MPPI optimization budget | baseline reward representations on FrankaKitchen · FrankaKitchen | aggregate success rate | Hurt |
| Use VIP as both reward and representation instead of VIP representation with sparse reward (ablation) | online RL on FrankaKitchen · FrankaKitchen | task success: 40% | Helped |
Closest papers by meaning
- VIP: Vision Instructed Pre-training for Robotic Manipulation (ICML 2025 · similarity 0.89)
- Robots Pre-train Robots: Manipulation-Centric Robotic Representation from Large-Scale Robot Datasets (ICLR 2025 · similarity 0.89)
- What Matters to You? Towards Visual Representation Alignment for Robot Learning (ICLR 2024 · similarity 0.88)
- Subtask-Aware Visual Reward Learning from Segmented Demonstrations (ICLR 2025 · similarity 0.88)
- Vision Language Models are In-Context Value Learners (ICLR 2025 · similarity 0.88)
- Video Prediction Policy: A Generalist Robot Policy with Predictive Visual Representations (ICML 2025 · 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.