Research map / Vision-language and multimodal / Robot action and navigation
MomaGraph: State-Aware Unified Scene Graphs with Vision-Language Models for Embodied Task Planning
Yuanchen Ju, Yongyuan Liang, Yen-Jen Wang, Gireesh Nandiraju, Yuanliang Ju, Seungjae Lee, Qiao Gu, Elvis Hsieh, Furong Huang, Koushil Sreenath
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 Graph-then-Plan (generate scene graph before planning) (ablation) | MomaGraph-R1 · MomaGraph-Bench overall | accuracy (%): 65.1 -> 71.6 | Improved |
| add Graph-then-Plan (ablation) | GPT-5 · MomaGraph-Bench overall | accuracy (%): 66.5 -> 71.6 | Improved |
| add Graph-then-Plan (ablation) | InstructBLIP-7B · MomaGraph-Bench overall | accuracy (%): 39.0 -> 39.5 | No clear effect |
| add Graph-then-Plan (ablation) | InternVL2.5-8B · MomaGraph-Bench Tier 1 | accuracy (%): 53.6 -> 51.0 | Hurt |
| use unified spatial-functional edges instead of spatial-only (ablation) | MomaGraph-R1 · MomaGraph-Bench overall | accuracy (%): 59.9 -> 71.6 | Improved |
| use unified edges instead of functional-only (ablation) | MomaGraph-R1 · MomaGraph-Bench overall | accuracy (%): 64.9 -> 71.6 | Improved |
| use unified edges instead of spatial-only (ablation) | LLaVA-Onevision · MomaGraph-Bench overall | accuracy (%): 54.0 -> 66.0 | Improved |
| train with RL (DAPO, graph reward) instead of SFT (ablation) | Qwen2.5-VL-7B · MomaGraph-Bench overall | accuracy (%): 63.9 -> 71.6 | Improved |
| train with RL instead of SFT (ablation) | Qwen2.5-VL-7B · BLINK | accuracy (%): 60.4 -> 63.5 | Improved |
| RL training instead of in-context graph examples only (ablation) | Qwen2.5-VL-7B · MomaGraph-Bench overall | accuracy (%): 60.2 -> 71.6 | Improved |
| vary reward weights (w_a, w_f, w_l) (ablation) | MomaGraph-R1 · MomaGraph-Bench overall | accuracy (%): 68.2 to 71.6 | No clear effect |
| RL-trained scene graph model on visual correspondence | Qwen2.5-VL-7B · BLINK | accuracy (%): 58.7 -> 63.5 | Improved |
Closest papers by meaning
- ManiTaskGen: A Comprehensive Task Generator for Benchmarking and Improving Vision-Language Agents on Embodied Decision-Making (NeurIPS 2026 · similarity 0.90)
- TaskGround: Structured Executable Task Inference for Full-Scene Household Reasoning (NeurIPS 2026 · similarity 0.88)
- OntoPlan: An Ontology-Grounded Scene Representation and Agentic Framework for Scalable Robot Task Planning (NeurIPS 2026 · similarity 0.88)
- Learning to Evolve Scenes: Reasoning about Human Activities with Scene Graphs (NeurIPS 2026 · similarity 0.87)
- PARTNR: A Benchmark for Planning and Reasoning in Embodied Multi-agent Tasks (ICLR 2025 · similarity 0.87)
- ESCA: Contextualizing Embodied Agents via Scene-Graph Generation (NeurIPS 2025 · 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.
