This research introduces a semi-supervised learning approach that reduces dependency on costly labeled data for vision-based language models in autonomous driving.

• Employs template-based prompts to extract valuable scene information from unlabeled driving data
• Significantly reduces annotation costs while maintaining or improving model performance
• Creates a more scalable approach to autonomous vehicle training
• Demonstrates how natural language understanding can enhance driving intelligence

For engineering teams, this represents a breakthrough in resource efficiency for autonomous systems development, potentially accelerating deployment timelines while maintaining safety standards.

Unlock the Power of Unlabeled Data in Language Driving Model