Evaluating VLMs' Spatial Reasoning Over Robot Motion: A Step Towards Robot Planning with Motion Preferences

Researchers evaluated four state-of-the-art Vision-Language Models (VLMs) on their ability to perform spatial reasoning for robot motion planning. Qwen2.5-VL achieved the highest performance at 71.4% accuracy zero-shot and 75% after fine-tuning, while GPT-4o showed lower performance in handling motion preferences and spatial constraints.

• →Qwen2.5-VL outperformed GPT-4o in spatial reasoning tasks for robot motion planning with 71.4% zero-shot accuracy.
• →Fine-tuning improved performance to 75% on smaller models, showing potential for optimization.
• →The study evaluated two types of motion preferences: object-proximity and path-style constraints.
• →Current VLMs show promise but still have limitations in enforcing complex spatial reasoning for robotics applications.
• →Research highlights the trade-off between accuracy and computational cost in token usage for VLM-robot integration.