Autonomous car company Argo AI just teamed up with The League of American Bicyclists (LAB) to set new guidelines for how self-driving cars should interact with cyclists. Argo AI hopes that other car companies will adopt these guidelines to improve road safety, though to be frank, we’re surprised that manufacturers haven’t already adopted such rules.
We don’t know exactly how many cyclists are injured or killed in car accidents, as these incidents aren’t always reported to the authorities. But the WHO states that around 41,000 cyclists are killed by cars each year, and at least 15,000 U.S. cyclists are hurt or killed in car accidents annually.
Clearly, we need more bike lanes and other forms of cycling infrastructure. But we also need to prepare for the future. Tesla and other vehicle manufacturers are prematurely pushing their self-driving tech into the real world, and this technology isn’t ready to deal with cyclists. That’s why Argo AI and LAB are setting new guidelines for self-driving cars and encouraging “rival” companies to do the same.
Here’s a paraphrased version of the guidelines drawn by Argo AI and The League of American Bicyclists:
- Make Cyclists an Object Class: To accurately detect cyclists, they should be recognized as a unique “object” separate from pedestrians or even scooters.
- Cars Should Anticipate Cyclist Behavior: Will a cyclist jump from the sidewalk to the road? Will they avoid obstacles in the street? To prepare for potentially dangerous circumstances, autonomous cars should predict every possible movement from a cyclist.
- Cycling Infrastructure Should Be Mapped: Autonomous cars should recognize bike lanes and other cycling infrastructure to avoid an accident. Additionally, they should know local cycling laws, which often vary from state to state. (In some states, cyclists can simply yield at a red light instead of waiting for it to turn green.)
- Cars Should Be Predictable to Cyclists: To mitigate risk, self-driving cars should act in a natural and predictable fashion whenever on the road. That means using turn signals and small movements to suggest intent.
- Cars Should Slow In Uncertain Situations: Self-driving cars tend to lower their speed and increase their distance from other vehicles when in “uncertain” situations. They should do the same when a cyclist is around, especially if the cyclists’ behavior is unpredictable.
- Cyclist Scenarios Should Be Tested Continuously: Manufacturers and engineers should continually test self-driving scenarios in virtual and real-world environments.
Most of these guidelines revolve around the idea that bicyclists are a unique and unpredictable obstacle for self-driving vehicles—which is true! Cyclists are faster than pedestrians and may weave in and out of the road to avoid parked cars or reach their destination more quickly. Autonomous cars should account for this behavior to avoid accidents, and if manufacturers are successful, they could dramatically the number of bicyclists that die in car accidents each year.
Manufacturers should feel compelled to adopt these guidelines, both for moral and business reasons. But it seems that vehicle manufacturers are more interested in selling self-driving tech than perfecting it. And that’s a big problem. Consumers and politicians aren’t very confident in this technology and will treat accidents, even if they are rare, as a sign that self-driving tech isn’t ready for the real world (which it may not be—Argo wrote these guidelines for a reason).
We hope that self-driving vehicle manufacturers will commit to Argo AI’s new guidelines or write their own safety rules to protect cyclists. Self-driving cars could dramatically improve road safety, but only if companies take steps to perfect autonomous driving technology and account for every possible obstacle.