Hit enter to search or ESC to close

Argo AI and the League of American Bicyclists Create Guidelines to Help Autonomous Vehicles Safely Coexist With Cyclists

Throughout its history, the nearly 142-year-old nonprofit organization League of American Bicyclists has advocated for the safety of the nation’s cyclists. 

“There’s so much conflict on our roadways,” says Ken McLeod, policy director at the League of American Bicyclists, “A lot of that is because of the way our roadways are designed, but also because of how people are using their vehicles. But I’m very hopeful that autonomous vehicles can fix some of these issues.” 

Through its collaboration with Argo AI, a leading self-driving technology company founded in 2016, the two organizations have shared a set of six technical guidelines that all autonomous vehicles should follow when operating on roads in order to protect cyclist safety. 

Why Autonomous Vehicles Must Proactively Address Cyclist Safety

The League was founded around the time that the first automobiles became available in the U.S. and it witnessed firsthand how cars became a fixture of American life, as well as the dangers that cyclists can face when riding alongside them. As part of its advocacy mission, the League gathers and analyzes statistics on cyclist and pedestrian fatalities and collects information on local bike and traffic laws from across the country. 

The data on traffic collisions in the U.S. shows just how far we need to go to improve cyclist safety in this country: there were 846 cycling fatalities recorded in the U.S. in 2019 alone — the latest year for which data from the National Highway Transportation Safety Administration (NHTSA) is available. 

The proportion of people killed in traffic collisions who were outside of a vehicle, which includes cyclists, pedestrians and scooter riders, is up to its highest recorded level at 34 percent, according to NHTSA, as the proportion of people killed inside of vehicles during collisions has fallen. 

New Self-Driving Guidelines Offer a Path Forward

The hope McLeod and the League express, that autonomous driving technology will lead to safer interactions between cyclists and vehicles, is one that Argo shares — going back to a set of cyclist and pedestrian safety principles established by Argo.

Argo’s consistent commitment to developing self-driving technology that is safe for pedestrians, cyclists and other road users, is exactly why the League partnered with Argo on the new technical guidelines.  

The goal for these guidelines? To provide a framework for the development and deployment of self-driving technology around cyclists, and what cyclists and members of the public should expect from self-driving vehicles. 

Argo’s SDS follows these cyclist and pedestrian guidelines as the company tests its vehicles on public roads in eight cities in the U.S. and Germany.

Both Argo and the League invite other self-driving tech companies to use them when designing and deploying their own systems. In order to protect cyclists, a SDS should abide by the following guidelines: 

  • Cyclists Should Be Recognized as a Distinct Object Class
  • Typical Cyclist Behavior Should Be Expected
  • Cycling Infrastructure and Local Laws Should Be Mapped
  • The Self-Driving System should have consistent and understandable behavior near cyclists.
  • The SDS should Prepare for Uncertain Situations and Proactively Slow Down
  • Cyclist Scenarios Should Be Tested Continuously

Each of the six technical guidelines entails a variety of bike-specific software design, testing, and deployment. Read on for some of the ways in which Argo develops self-driving tech in accordance with the guidelines.

Establishing Cyclists as a Distinct Object Class 

How does a self-driving car recognize what a cyclist is and distinguish it from all the many other objects the use the roadways? To begin, Argo labelers analyze video footage and still imagery of roadway environments collected by the cameras on Argo’s self-driving test cars. They label the important objects that appear in the footage, including other vehicles, pedestrians, and cyclists.

“We have images of cyclists from all different viewpoints and engaged in all different kinds of behaviors,” says Andrew Hartnett, Argo’s Tech Lead Manager of Deep Forecasting. “They’re gathered primarily from our own cars, which provides several benefits: they match the distribution of cyclists that we expect to see in the cities we operate, the camera angles match, and it’s easier to train the system on the same type of images you will be seeing when driving in the real world.” 

These images are used to train and test Argo’s deep perception models, which allow the SDS to recognize the tell-tale signs of cyclists, such as spoked wheels, thin metal frames, riders standing or sitting, an object that changes position and is not fixed or stationary, and so on. This even works for more uncommon cycle models, such as recumbent bicycles and unicycles, because Argo has examples of these in its training data 

The combination of the labels and the SDS’s prediction capabilities is what helps the SDS establish cyclists as a “distinct object class.” In essence, it’s a way for Argo’s SDS to know that if it sees something exhibiting any number of typical cyclist visual qualities and/or cycling behaviors, the SDS should treat what it’s seeing as a cyclist. 

Expecting Typical Cyclist Behavior

Once Argo’s SDS identifies a road user as a cyclist, it uses its prediction subsystem to anticipate the numerous ways the cyclist could behave in the real world — how fast it’s likely to travel, in which directions, on which side of the lane or sidewalk, based upon both that particular cyclist’s observed speed and trajectory, and its library of knowledge of cyclists it has seen in the past. 

While each cyclist will behave uniquely to some extent based on other vehicles and road conditions, cyclists as a whole exhibit certain behaviors that are not seen as often in pedestrians or motor vehicles. 

“Cyclists sometimes do things like yield at a stop sign rather than come to a complete stop,” McLeod says, by way of an example. 

This particular behavior, sometimes nicknamed an “Idaho Stop,” after a 1982 state law that permitted the move for cyclists only, is distinctly different from the legal behaviors of car drivers in the same situation. As a consequence, the Idaho Stop can frustrate human drivers. 

As McLeod explains: “If you don’t think of cyclists as individuals that have distinct behaviors, and if you decide that they should follow the same exact laws as vehicles rather than understanding the way they actually behave on the roads, that’s what can create conflict. Cyclists have distinct abilities from drivers of cars and recognizing that can reduce conflicts on our roads and in our culture.” 

Another distinct behavior that cyclists sometimes exhibit is swerving suddenly away from the side of the lane where they’re riding, to avoid being hit by people opening doors of cars parked along the streets.  

These, and numerous other distinctive cyclist behaviors, are all actions a safely designed SDS needs to anticipate and expect, and Argo’s SDS is developed to handle.  

Mapping Local Cyclist Infrastructure and Knowing Local Laws

Every Argo self-driving test vehicle driving on public roads is loaded with a set of constantly updated 3D maps of its local driving area.

These maps contain traffic signals, lane markings, lane directions, construction, even some vegetation and any other information — including bicycle lane markings — that is highly important for the SDS’s ability to predict a cyclist’s movements. 

More than this, Argo’s fleet adjusts to local driving laws and customs. As Argo’s 2021 Safety Report specifies, depending on the situation: “A car may occupy the bike lane when turning right on red, or in Pennsylvania, where the same right-turning car must remain in the vehicle lane, the SDS ensures that it follows the rules in the jurisdiction in which it operates.” 

Finally, Argo’s SDS does more than just recognize local biking infrastructure and laws and drive accordingly, giving the right-of-way to cyclists: it also anticipates where cyclists are likely to ride on a road even when there is no designated bike lane. If a cyclist is riding on the right side of an unmarked lane, Argo’s SDS anticipates that it will follow this path even when crossing through an intersection or when turning — unless there is some obstruction or lane marking, or cyclist behavior to suggest otherwise. Argo’s software proactively identifies where a cyclist is riding, and gives them appropriate space, creating a bike lane of sorts even where there may be none. 

Creating Consistent and Understandable Self-Driving Behavior

Designing a SDS that recognizes cyclists and expects common cyclist behavior is only one part of the equation for keeping cyclists safe. The other part falls on self-driving cars to display and exhibit consistently safe, predictable driving behavior around cyclists, so that cyclists and other road users can anticipate and trust the SDS. 

For the Argo SDS, this task falls primarily upon a few distinct subsystems: the prediction system and the motion planning and control system.

The prediction system anticipates what the other road users around a self-driving car will do next, forecasting a wide range of possible paths they could take and assigning each a probability score. 

The motion planning and control system uses this information to decide where the self-driving car should move in relation to the other users and objects to keep everyone safe. 

The SDS will consistently provide cyclists the space they need given their relative speed and road conditions. To achieve this, Argo vehicles follow certain parameters for appropriate lateral and following distance. For an Argo vehicle traveling at about 30 mph, that means striving for 4-5 feet of lateral distance when passing a cyclist and about 70 feet when following a cyclist.

Both of these subsystems work together to create a safe, naturalistic driving experience that cyclists and other road users can rely on. 

Preparing for the Uncertain and Slowing Down

Even with the aid of an advanced motion forecasting and control system, the presence and actions of cyclists can be unpredictable. Cyclists often ride on sidewalks, walk the bike through crosswalks, or even travel in the opposite direction of oncoming traffic — and can transition between these modes quickly. 

Furthermore, due to the relatively smaller size of bikes compared to cars, cyclists can become occluded on the roadways by other cars and trucks, and may disappear and reappear suddenly. 

To safely navigate this, the Argo vehicle will proactively slow down in situations where the cyclist’s future behavior is uncertain. This includes cyclists approaching an intersection, cyclists deviating from traveling parallel to the roadway, and cyclists in the vicinity of crosswalks and the start/end of cycle infrastructure like bike lanes.

Additionally, Argo’s self-driving system’s suite of overlapping sensors — namely, cameras, radar, and lidar — detect objects 360º around the vehicle. When the SDS encounters something unusual that its sensors cannot definitively help classify, it treats it with additional caution and proactively slows down. 

Continuous Testing 

Argo tests its SDS constantly to ensure it recognizes cyclists and drives safely around them in all the situations in which they may appear — including rare and unusual situations, such as a fleet of unicycles appearing suddenly on the roads. 

This is done through a variety of testing processes, beginning with running every new software update to its SDS through virtual simulations. Argo engineers use simulation testing to adjust different variables — the number of other vehicles on the road, pedestrians, cyclists, for some examples — as well as prepare for rare or unseen situations, to see how the software responds. 

But Argo doesn’t stop there. Once a software update passes simulation testing, it is installed on a real car and taken to Argo’s closed course test track in Western Pennsylvania. At the test track, Argo’s Structured Testing team creates real-world cyclist props to ride alongside real self-driving test vehicles. They also incorporate trained test cyclists to ride alongside the cars. 

Once the SDS demonstrates it is reliably and safely driving alongside cyclists on the test track, it is implemented in Argo vehicles operating on public roads, where it is tested amongst city traffic under the supervision of trained Test Specialists, or safety operators.

How Autonomous Vehicles Can Help Cyclists

Argo’s approach to developing, testing and deploying autonomous vehicles follows the framework of the new technical guidelines, and gives the League the hope that self-driving cars will help to usher in a new era of safer shared streets between cars and cyclists.

Before the release of the six new guidelines, McLeod saw Argo’s headquarters and autonomous vehicle technology up-close and rode as a passenger in a test vehicle on the company’s track in Western Pennsylvania, as well as on the public roads of Pittsurgh. His confidence in Argo’s SDS was bolstered by what he saw and experienced. 

“My ride with Argo in Pittsburgh was a little boring, which was a good thing,” McLeod says with a laugh. “By that I mean, the autonomous vehicle was so safe and confident. It felt like a good, experienced driver, which was exactly what I wanted to see.” 

McLeod has a personal investment in making sure self-driving technology continues to improve how cars share the road with cyclists. 

“As a young person, bicycles were really important to my mobility,” McLeod says. “And we want to ensure that opportunity continues safely for all young people, and people of all ages, going forward. Designing responsible autonomous vehicles is one important step for doing that.”

Choose your lane

How Autonomous Vehicles Distinguish Between Bicycles and People Who Ride Them

How Autonomous Vehicles Distinguish Between Bikes and People

When it comes to how autonomous vehicles see the world, humans come first, literally. Autonomous vehicles (AVs), like the kind operated by Pittsburgh-based Argo AI, use Machine Learning to detect and classify the objects in their surroundings, identifying people...
Why The League of American Bicyclists is optimistic about autonomous vehicles

Why a Leading Cycling Advocacy Group Is Optimistic About Autonomous Vehicles

As autonomous vehicle use grows, AV companies and the League of American Bicyclists are collaborating on how to ensure cyclists and motorists can share the roads safely, even if the “motorist” is artificial intelligence software. As part of the...

Self-Driving Is Arriving Right On Time. Just Like Ice Cream Did

Seven years ago, I was a self-driving skeptic. Not of the technology. Of all the “experts” promising autonomous vehicles would be everywhere by 2020. You didn’t need to be Nostradamus to know that was ridiculous. All you needed was...
Illustration of a futuristic parking deck turned into a mixed-use space, with AVs driving by

How Autonomous Vehicles Could Help Transform Parking Lots

Researchers say it’s likely that autonomous vehicles (AVs) can help reduce the need for parking lots, opening more room for grass and trees and other elements of nature. It may not seem like it when you’re circling the block...
An illustration of an Argo autonomous vehicle in teal against a backdrop of buildings, a bicyclist, and research papers

7 Big Breakthroughs From Argo Research at CVPR 2022

The 2022 Conference on Computer Vision and Pattern Recognition (CVPR 2022) is nearly here. Thousands of computer scientists, software engineers, and researchers from around the globe will gather in New Orleans to review and discuss their latest work in...

Researchers Predict the Future With Lidar Data

Researchers at Carnegie Mellon University’s Argo AI Center for Autonomous Vehicle Research, a private-public partnership funded by Argo for advancing the autonomous-vehicle (AV) field, say they have come up with a way to use lidar data to visualize not...

Must Reads