OTTER introduces a novel Vision-Language-Action architecture that helps robots better understand instructions and visual environments for improved task performance.

• Maintains semantic alignments from pre-trained vision-language models through text-aware visual feature extraction
• Processes only task-relevant visual elements instead of entire scenes
• Demonstrates superior performance in robotic instruction following tasks
• Provides a more efficient approach than existing models requiring extensive fine-tuning

This research significantly advances robotic engineering by creating more intuitive human-robot interfaces and improving robots' ability to operate in complex, real-world environments with natural language instructions.

OTTER: A Vision-Language-Action Model with Text-Aware Visual Feature Extraction