Research map / Vision-language and multimodal / Robot action and navigation
LIV: Language-Image Representations and Rewards for Robotic Control
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 |
|---|---|---|---|
| fine-tune LIV-EPIC with the LIV objective | LIV-EPIC reward model · MetaWorld | success rate: 29.7 -> 55.2 | Helped |
| fine-tune LIV-EPIC with the LIV objective | LIV-EPIC reward model · FrankaKitchen | success rate: 1.3 -> 20.0 | Helped |
| use LIV-EPIC rather than the strongest listed baseline representation | language-conditioned behavior cloning · FrankaKitchen | success rate: 22 ± { 3.5} -> 29.3 ± 4.6 | Helped |
| use LIV-EPIC rather than the strongest listed baseline representation | language-conditioned behavior cloning · MetaWorld | success rate: 24.2 ± { 3.0} -> 30.6 ± { 5.0} | Helped |
| use language task encoding rather than one-hot task IDs (ablation) | LIV-EPIC policy · FrankaKitchen | success rate: 17.6 ± { 5.0} -> 29.3 ± 4.6 | Helped |
| use language task encoding rather than one-hot task IDs (ablation) | LIV-EPIC policy · MetaWorld | success rate: 26.1 ± { 5.5} -> 30.6 ± { 5.0} | Helped |
| fine-tune CLIP with LIV rather than CLIP alone (ablation) | CLIP representation · MetaWorld | success rate: 16.4 ± { 4.3} -> 33.9 ± { 7.5} | Helped |
| fine-tune CLIP with LIV rather than CLIP alone (ablation) | CLIP representation · FrankaKitchen | success rate: 14.0 ± { 6.8} -> 26.8 ± { 4.9} | Helped |
| replace LIV fine-tuning with CLIP-only fine-tuning (ablation) | CLIP representation · MetaWorld | success rate: 33.9 ± { 7.5} -> 16.4 ± { 4.3} | Hurt |
| replace LIV fine-tuning with CLIP-only fine-tuning (ablation) | CLIP representation · FrankaKitchen | success rate: 26.8 ± { 4.9} -> 14.0 ± { 6.8} | Hurt |
| remove the semantic InfoNCE term and fine-tune with VIP-I only (ablation) | CLIP representation · FrankaKitchen | success rate: 26.8 ± { 4.9} -> 14.8 ± { 1.3} | Hurt |
| fine-tune CLIP with LIV instead of TCN+CLIP | CLIP representation · FrankaKitchen, 10 demonstrations per task | success rate: 13.3 ± { 2.3} -> 22.0 ± { 6.0} | Helped |
Closest papers by meaning
- Vision Language Models are In-Context Value Learners (ICLR 2025 · similarity 0.89)
- NS-VLA: Towards Neuro-Symbolic Vision-Language-Action Models (NeurIPS 2026 · similarity 0.89)
- GoalLadder: Incremental Goal Discovery with Vision-Language Models (NeurIPS 2025 · similarity 0.89)
- Vision-Language Models are Zero-Shot Reward Models for Reinforcement Learning (ICLR 2024 · similarity 0.88)
- SOLE-R1: Video-Language Reasoning as the Sole Reward for On-Robot Reinforcement Learning (NeurIPS 2026 · similarity 0.88)
- ELEMENTAL: Interactive Learning from Demonstrations and Vision-Language Models for Reward Design in Robotics (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.