How Will Driverless Cars Share Roads with Cyclists?
According to a recent Bloomberg report, over 25 percent of cars sold will be fully driverless by 2035. For now, there's still a long way to go, as developers overcome certain challenges that autonomous vehicles will encounter on public roads, such as snow and fickle-minded cyclists.
Focusing on the latter concern, pedestrians on bikes were one of the first issues developers raised at the very early stages of testing. Today, that problem has not been solved. Google (before Waymo) mentioned in its June 2016 report that movements of cyclists are difficult to predict. According to the tech giant, over 50,000 individuals riding bikes were injured in 2016.
Deploying the Deep3DBox Algorithm
The movement of cyclists are difficult to predict because they are spontaneous. As a result, Google's self-driving platform is programed to provide ample space for them to move freely within their lanes. One of the ways autonomous cars detect the movement of people on bikes is through hand signals. Developers have noted that cyclists tend to signal very early before making the actual turn. Hence, when a sensor reads a hand signal, it must "remember" it for some time.
Fast-moving autonomous cars can't just rely on hand signals, simply because not all cyclists use them consistently. Furthermore, a wave at a friend at the corner of the street or wiping sweat off one's forehead could easily generate a false reading.
At the very basic level of detection, driverless sensors will have to first know which way the bicycle is facing. To streamline detection, developers are turning to the Deep3DBox algorithm. When deployed, it predicts which way a car or bicycle is facing. So far, this option is one of the most effective ways to detect cyclists. During trials, the robust algorithm successfully spotted an individual on a bike 74 percent of the time. Furthermore, it correctly detected the direction cyclists were facing roughly 59 percent of the time.
"Deep learning is typically used for just detecting pixel patterns. We figured out an effective way to use the same techniques to estimate geometrical quantities," explained Deep3DBox contributor Jana Košecká, a computer scientist at George Mason University in Fairfax, Virginia.
Network Sensors and Expanding Bicycle Lanes
There are several solutions to this issue, with some not coming from developers of autonomous vehicles. A handful of countries, including Germany, are expanding bike lanes to cater to cyclists, so they don't have to fight with drivers for space on public roads. This solution may also help reduce traffic in urban locations, allowing vehicles to travel at a more consistent pace.
A more modern solution involves installing a network of sensors around roads and high-activity areas. The sensors could detect cyclists from a different POV and relay that information to nearby driverless cars.
"A car is basically a big block of stuff. A bicycle has much less mass and also there can be more variation in appearance — there are more shapes and colors and people hang stuff on them," said Nuno Vasconcelos, a visual computing expert at the University of California, San Diego.
- Applied Intuition Develops Virtual Simulator for Driverless Cars, Completes Series A Round
- US Department of Energy Issues $2M Grant to Develop Autonomous Fueling Stations
- Scientists Develop Hazardous Gas-detecting Drones, Receives Grant from the NSF
- France Partners with Hitachi and Bosch to Launch Autonomous Trains
- Can More Home-based Charging Options Boost EV Adoption?
- Volvo Showcases Autonomous, All-electric 360c Driverless Concept Car
- Fiat Chrysler to Open New Driverless Test Site, Invests $30 Million
- Auto Startup HEVO Wants to Make EV Charging Wireless