From Idea to Europe: Research Journey on AI and Guide Dogs

This September, I’ll be presenting my first peer-reviewed paper at the AAATE 2025 Conference at the European University of Cyprus. My research — conducted over two years — is about developing an AI-powered assistive technology for the visually impaired. It uses machine learning to detect when a guide dog is becoming distracted, in real-time.

I began this journey with a simple question: Can AI help guide dogs stay focused, and make life safer for the visually impaired?

The Big Idea

Guide dogs are incredible — they don’t just detect obstacles like a white cane, they actively avoid them. But like any living creature, they get distracted — by other dogs, people, sounds. These moments can be dangerous for blind handlers who might not even realize their dog has lost focus.

So I built a system that:

• Uses DeepLabCut (a pose estimation model) to track dog body posture from video

• Trains an LSTM (Long Short-Term Memory) neural network to recognize distraction behaviors

• Runs in real-time, so it could be used on a chest-mounted or glasses-mounted device with a haptic wristband that vibrates when the dog is distracted

How I Built It

• Conducted expert interviews with top guide dog trainers (from Guiding Eyes and Guide Dogs for the Blind)

• Analyzed over 50 videos and 100,000+ frames of guide dog footage

• Labeled over 1,700 distraction events manually

• Trained and tested an LSTM-based machine learning model using TensorFlow

• Achieved 87% precision and 82% F1-score in detecting real distraction events

What Makes It Unique

Most assistive technologies focus on helping blind users understand their environment. This system enhances their relationship with the guide dog — a trust-based, deeply personal connection. It’s the first project (to my knowledge) that targets real-time distraction detection in service animals.

Results That Matter

The model worked. It predicted distractions with high accuracy — and improved rapidly as it was trained with more data. The potential benefits are huge:

• Safer, more confident mobility for the visually impaired

• Reduced training time and better feedback for guide dog trainers

• Broader applications for any service animal or safety-critical partnership

What’s Next

This fall, I’ll present my paper in Nicosia, Cyprus at the AAATE 2025 Conference, which brings together leaders in assistive technology from around the world. I’m also exploring ways to turn this into a real product — possibly even a startup.

Stay tuned for updates — I’ll share more behind-the-scenes on how I built the system, what I learned at the conference, and where I hope to take this next.

Call to Action

Curious about how machine learning can improve human-animal collaboration? Or want to see how AI can create social good? Drop a comment or subscribe — I’ll be sharing all the details soon.