Purple tape is stretched in a straight line across a large area surrounded by trees. A woman with a smartphone strapped to her waist and a relay race number with the number 6 on it is running along the line of the tape.

Project Guideline: Towards helping everyone to run freely

Project Guideline is an early-stage research project exploring how Google AI can help people with visual impairments run freely by themselves.

In Japan, around 60 percent of the population exercises routinely.

However, that number drops to around 30 percent amongst people with visual impairments.

Enjoying sport and having the opportunity to participate is a right granted to everyone by law. However, this right is not always a reality for those with various disabilities. For example, the simple freedom to run alone is an almost impossible dream for those with visual impairments.

“Is there a way to use technology to enable runners with visual impairments to run by themselves?”

This question was posed to us by Thomas Panek, President and CEO of Guiding Eyes for the Blind and accomplished marathon runner in the United States, and it became our starting point. Everyone should be able to pursue their own physical potential freely and independently. We started this experiment as the first small step toward achieving such a society.

We believe in building products that work for everyone. Making use of now widely adopted smartphones and headphones, we built the first prototypes together with Thomas and announced Project Guideline in the United States in 2020.

As Parasports and sports in general gained more attention in 2021, we announced the next stage of the project in Japan. Masamitsu Misono, a technology advocate for the blind community, accomplished runner, and person who is blind totally, was our first partner.

Towards a world where everyone can pursue their fullest potential

In 2022, we got together with NPO Achilles International Japan, an organization that connects people with disabilities and able-bodied people to enjoy running and walking together. We helped visually impaired runners participate in the ASICS World Ekiden 2022, a virtual Ekiden race that connects digital sashes in teams of six, without any escort runners. As a result, all six runners ran their segment with the help of Project Guideline and completed the 42.195km in 4 hours 29 minutes 44 seconds. The team competed equally against other teams of able-bodied athletes from all over the world.

There are countless ways to enjoy running. You can run by yourself , run with a friend or compete with like-minded runners. It shouldn’t matter if a person has a disability or not. We had the opportunity to demonstrate how a team of visually-impaired runners could compete equally against teams of able-bodied athletes through a virtual race and with Project Guideline’s technology. The challenge highlights how Project Guideline has the potential to help everyone pursue their fullest potential.

How Project Guideline works

Project Guideline uses image recognition technology leveraging on-device machine learning that runs on Android smartphones. It identifies a colored line on the ground, determines whether the line is to the left, right or center of the runner, and sends audio signals through headphones worn by the runner. Based on that feedback, the runner can correct their running position to stay on the line and enjoy their run.
Animation showing how Project Guideline communicates to users through sound.
It’s easy for the human eye to identify a line on the road, but asking a machine to process that information is no easy task. The camera worn by the runner shakes constantly, and when outside, the direction and brightness of light changes continually. Shadows and leaves can cover the course, and even the color of the ground itself is not constant. You also have to address any curves in the course.
Animation showing how Project Guideline recognizes line directions and angles.
To make accurate decisions under these varying conditions, we developed a machine learning model that utilizes TensorFlow™, an open-source library for machine learning published by Google. We continue to work on improving the accuracy and performance of the system by gathering video data from as many situations as possible to train the model to recognize different environments. Most importantly, we recognize the importance of involving community experts in the design and engineering processes. That’s why we’re partnering with the leading organizations in the United States and Japan to get direct feedback from athletes who are blind to make the system safer and easier for users.

Partners

Project Guideline is an early-stage research project. We have been collaborating with partners to conduct events and field tests for data collection and user feedback.