CYBATHLON Challenges 2023

Vision Assistance Race

Overview

CYBATHLON, organized by ETH Zurich, challenges teams worldwide to develop assistive technologies for people with disabilities. I led team FlashLight in the Vision Assistance Race at CYBATHLON Challenges 2023 — the inaugural year for this discipline.


Tasks

The Vision Assistance Race consisted of two tasks designed for blind or visually impaired pilots:

Task 1: Sidewalk Navigation — Navigate to the end line without bumping into any obstacles.

Task 2: Serving Food — Pour water to a designated level, carry a balanced tray with bowl and glass, and place items without touching existing utensils.

Two tasks for the Vision Assistance Race in CYBATHLON Challenges 2023.

Technical Approach

Scene Understanding

We utilized YOLOv8 for object detection to understand the scene and determine navigation commands.

Haptic Feedback

Vibration motors provided real-time feedback to the pilot, conveying spatial information and navigation commands.

Target Location Detection

For Task 2, we detected the dining table bounding box and compared it with utensil positions to determine the correct placement direction.


Results

We placed 2nd out of 2 teams in the Vision Assistance Race. As the inaugural year for this discipline, participation was limited, but the experience provided valuable insights into developing assistive technology under competition constraints.

Watch the full competition video on the official CYBATHLON website.


Team

FlashLight team

Our team included pilots with visual impairments, a guide dog trainer, faculty members, and students developing hardware and software. Team name FlashLight — lighting up the path for blind or visually impaired individuals.


Takeaways

  • Competition solutions differ from real-world deployments — heuristic methods may work in controlled settings but lack robustness.
  • Simplicity is key — our multi-sensor system added complexity that was difficult to optimize under time constraints.
  • User feedback matters — pilots noted some tasks had limited real-world relevance.