Imagine how much safer a cyclist would be if they had eyes in the back of their head. No more panicked glances over the shoulder to check if a car is quickly approaching. Fewer near-collisions when swerving to avoid a car passing a little too closely.

They would simply see these hazards coming.

That's the goal of Copilot, a device designed by Carnegie Mellon School of Computer Science alumni that uses technology typically found in autonomous vehicles to watch the area behind a cyclist and alert them to oncoming traffic or other risks outside their field of vision.

"We see this as a way of alerting the driver as well as the cyclist," said Clark Haynes, co-founder of Velo.ai, the company developing Copilot. Haynes earned his Ph.D. in robotics from CMU and has been working in the robotics and autonomous vehicle industries in Pittsburgh since the early 2000s.

The Copilot prototype is about as big as a grapefruit, though Haynes says the actual device will be about a fifth of that size. It attaches to the back of a bicycle in the same location as a rear bike light. Using camera sensors similar to those found on a smartphone, Copilot views the field behind the bicycle and sends that information to an artificial intelligence chip that uses computer vision models to analyze potential threats and alert the cyclist. If something is approaching, like a car, Copilot emits a warning from its 90-decibel speaker along with a bright flashing light.

Haynes is a lifelong bike commuter, but stopped going to the office when the COVID-19 pandemic started. Instead, he started riding on trails and took to the city's streets when he needed new terrain to explore. 

"That's where I really encountered how aggressive drivers can be," he said.

Of the 24,698 vehicle crashes reported to the City of Pittsburgh Crash Data Dashboard between 2017 and 2022, 1,214 involved a pedestrian and 238 involved a bicycle. Four of the bicycle crashes resulted in deaths while 19 led to serious injury. Pedestrian crashes resulted in 28 deaths and 152 reports of severe injuries. Together, deaths resulting from pedestrian or cyclist crashes accounted for about 27% of all crash fatalities during that period.

Nationally, about 1,000 bicyclists die and more than 130,000 are injured each year, according to data compiled by the Centers for Disease Control and Prevention. Another analysis from the U.S. Department of Transportation National Highway Traffic Safety Administration shows that bicycle fatalities have been rising since 2011, when there were 682 deaths nationwide. In 2020, there were 938 deaths, accounting for 2.4% of all traffic crash-related fatalities, according to the most recent data available.

Haynes started thinking about the technology that goes into making autonomous vehicles safe — perception, prediction and risk estimation — and was inspired to find a way to put that technology into the hands of cyclists and pedestrians. 

"I think the challenge with autonomous vehicles is the long tail of issues," he said. "You can get a 99.9% good autonomous vehicle, but it's still not good enough because it can't do everything."

While the autonomous vehicle industry struggles to reach perfection, Velo.ai operates on the thesis that the same technology can be used in more focused ways to help a pedestrian or cyclist make their surroundings safer.

Motorists will recognize this trend in the bevy of assistive driving features such as adaptive cruise, automatic emergency braking, and lane departure correction available in newer vehicle models.

"People realize that there are things that have been developed that can be rolled out now and can make a difference in automotive safety," said Matthew Johnson-Roberson, director of the Robotics Institute. "Copilot is a great application of doing that same kind of thing, except for bicycle safety."

Johnson-Roberson notes that it's not surprising that this type of technology is coming out of Pittsburgh. As a cyclist who bikes in the city with his daughter, Johnson-Roberson pointed out that Pittsburgh has tough roads and nascent bike infrastructure — like protected bike lanes separating cyclists from vehicle traffic by a barrier or a parking lane — to keep everyone on the road safe. But the city also has a vibrant cycling community and a robust network of robotics, computer vision and artificial intelligence experts.

"That means there are people who understand how this technology works and would be willing to put it on their own bikes and give it a shot," Johnson-Roberson said.

That combination of experience and need makes Pittsburgh a natural testing ground for technology like Copilot, which started deploying prototypes on city streets in October 2022. As of this spring, about 30 users are testing Copilot in Pittsburgh and about 400 are on the waitlist to try the product.

"It surprised me. We had many of our testers say how useful it was," Haynes said. "Many of our testers felt more confident, and we're only more excited to make it better and better and really push the boundary of what the tech is capable of doing."

Moving forward, the Copilot team hopes to make the device more user friendly by improving things like battery life, compatibility with a smartphone app, and the ease of taking it on and off the bike.

On the technology side, Micol Marchetti-Bowick, co-founder of Velo.ai, says there's room to expand Copilot's predictive capabilities. Machine learning will help train Copilot to decide how urgent a threat might be and refine the alerts to ensure cyclists aren't receiving too many warnings.

Similar products on the market use radar technology to detect objects around a cyclist. But where those devices merely perceive present roadway hazards, Copilot goes a step further to predict what those hazards might do next — or determine if they're really a risk at all, Marchetti-Bowick said.

"A cyclist doesn't have time to pay attention to a pedestrian that's behind a parked car, that's in front of another car, that's in front of them. That's many layers of objects away," she said. "But a computer can pay attention to all of those things continually without getting tired."

Marchetti-Bowick earned her Ph.D. from CMU's Machine Learning Department. Her career intersected with robotics and self-driving vehicle technology, with much of her work focusing on prediction technologies similar to what has been adapted for Copilot.

"The goal is to make roads safer and reduce the number of vehicles we even need in the world," she said of working on autonomous vehicles. "I think the idea of having similar goals by focusing directly on bikes and micromobility, and cities too — thinking about how we can make urban spaces and roads safer and more welcoming to nondrivers — was what drew me to Velo."

The startup is currently conducting a pre-seed round and has received investment from the Richard King Mellon Foundation's social impact fund, as well as Reinforced Ventures. Copilot is scheduled to launch to the public this summer.