Research map / Vision-language and multimodal / Multimodal LLMs
CoSPlan: Corrective Sequential Planning via Scene Graph Incremental Updates
Shresth Grover, Priyank Pathak, Akash Kumar, Yogesh Rawat
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 single-shot scene graph with incremental scene graph updates (SGI) (ablation) | GPT-4o · CoSPlan Maze-E Step Completion | accuracy (%): 46.1 -> 56.1 | Improved |
| replace single-shot scene graph with SGI (ablation) | Intern-VLM 2 · CoSPlan Blocks-World-E Step Completion | accuracy (%): 18.9 -> 29.2 | Improved |
| replace single-shot scene graph with SGI (ablation) | GPT-4o · CoSPlan Error Detection (Robo-VQA-E) | accuracy (%): 44.2 -> 57.4 | Improved |
| replace single-shot scene graph with SGI (ablation) | Qwen2 VL-8B · CoSPlan Step Completion (all tasks) | accuracy (%): Maze-E 28.3 -> 28.5; Blocks-E 18.8 -> 18.5 | No clear effect |
| add scene graph over CoT prompting (ablation) | GPT-4o · CoSPlan Error Detection (Maze-E) | accuracy (%): 40.3 -> 35.3 | Hurt |
| use ground-truth (oracle) scene graph instead of VLM-generated (ablation) | SGI on Intern-2 · CoSPlan Maze-E Step Completion | accuracy (%): 43.2 -> 59.0 | Improved |
| perturb scene graph object positions (noisy SG) (ablation) | SGI on Intern-2 · CoSPlan Blocks-World-E Step Completion | accuracy (%): 29.2 -> 25.4 | Hurt |
| increase number of errors in the context (ablation) | InternVLM (SG) · Maze-E Step Completion | accuracy (%): 41.2 -> 22.5 (1 -> 7 errors) | Hurt |
| apply SGI to VQA by iterating over MCQ options | Intern-VLM · Spatial-Map VQA | accuracy (%): 41.3 -> 44.3 | Improved |
| apply SGI instead of SG on VQA | CoG-VLM · Spatial-Map VQA | accuracy (%): 36.7 -> 35.8 | Hurt |
| add scene graph over vanilla prompt for open-source VLMs (ablation) | Janus-pro-7B · CoSPlan Robo-VQA-E Step Completion | accuracy (%): 14.1 -> 21.3 | Improved |
| increase initial context length (more performed steps) (ablation) | VLMs on CoSPlan · CoSPlan Step Completion | accuracy (%) | Improved |
Closest papers by meaning
- Visual Planning: Let's Think Only with Images (ICLR 2026 · similarity 0.89)
- CogCoM: A Visual Language Model with Chain-of-Manipulations Reasoning (ICLR 2025 · similarity 0.88)
- AdaReasoner: Dynamic Tool Orchestration for Iterative Visual Reasoning (ICLR 2026 · similarity 0.88)
- InfiniBench: Infinite Benchmarking for Visual Spatial Reasoning with Customizable Scene Complexity (CVPR 2026 · similarity 0.87)
- MetaSpatial: Reinforcing 3D Spatial Reasoning in VLMs for the Metaverse (ICLR 2026 · similarity 0.87)
- Evaluating Vision-Language Models as Evaluators in Path Planning (CVPR 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.
