CANINE: AI Coaching System for Robot Guide Dog Training
Researchers have developed CANINE, an automated coaching system that trains visually impaired users to navigate with a robot guide dog through personalized verbal feedback. The system decomposes complex coordination tasks into sub-skills and operates at two levels: high-level tracking of learner proficiency across sub-skills using knowledge tracing to prioritize weak areas, and low-level observation of practice episodes to infer error causes via foundation models and generate targeted corrections. A controlled study demonstrated its effectiveness. The work addresses the challenge that effective human-robot coordination is difficult to learn from generic instructions alone.
Key facts
- CANINE is an automated coaching system for robot guide dog navigation.
- It targets visually impaired users.
- The system uses personalized, adaptive verbal feedback.
- It decomposes coordination tasks into sub-skills.
- High-level training decisions use knowledge tracing to prioritize weak areas.
- Low-level training uses foundation models to infer error causes.
- The system generates targeted verbal corrections.
- A controlled study was conducted to evaluate CANINE.
Entities
—