You hear it before you see it on a narrow road. A car hangs behind you, not passing, and you wonder if the driver has seen you or is staring at a phone.
A car with nobody behind the wheel sounds worse. That empty seat feels like one more thing you can’t predict.
So it’s fair to ask whether self-driving cars are safe for cyclists in real traffic. Waymo robotaxis carried public riders in 15 US cities by September 2026, and parts of the same sensing tech are already in ordinary new cars as automatic braking.
The numbers, the independent check, what went wrong and what it means on your roads are all below, including a twist.
What Waymo’s Own Crash Data Says About Cyclists
Nine crashes is both encouraging and sobering.
Through June 2026, Waymo reports 86% fewer injury crashes involving cyclists than average human drivers over the same distance.
That figure sits in Waymo’s safety hub, based on about 271 million rider-only miles. Rider-only means no human driver on board, just the car driving itself with passengers.
The comparison covers Phoenix, the San Francisco Bay Area, Los Angeles, Austin and Atlanta. It counts crashes regardless of fault and only on the road types Waymo uses, with freeways excluded.
The count was 9 Waymo cyclist injury crashes against about 65 expected for humans. Nine still happened.
Think of a student marking their own homework and handing in good grades. The method looks serious, with the human figure adjusted for crashes that never get reported to police. It still needs an outside examiner.
The League of American Bicyclists supports self-driving development. The League argues these cars should eventually make riding safer because they can’t get distracted, and the League names Waymo among the companies endorsing its cyclist-safety efforts on self-driving cars. Intent is not proof, and that’s where an outside examiner comes in.
What the Independent Crash Study Found
Three crashes cannot settle an argument, even when the rest of the report looks strong.
In July 2026, the Insurance Institute for Highway Safety published an independent study of Waymo crash rates. Across about 50 million driverless miles from 2021 through 2024, with freeways excluded, Waymo had 68% lower police-reportable crash rates than human drivers in the same cities and years.
David Harkey, president of the Insurance Institute for Highway Safety, put it carefully in the IIHS announcement: “on a limited scale, these driverless cars are safer than human drivers.” He added that the present data collection system isn’t good enough to allow continuous monitoring of a large-scale expansion.
For cyclists specifically, the IIHS paper could not confirm Waymo’s benefit. Crashes per million miles means crashes divided by miles driven, scaled up so rare events can be compared. With only three Waymo cyclist crashes, the math can’t show a difference either way.
| Waymo safety hub | Independent IIHS study | |
|---|---|---|
| Who counted | Waymo, the company | Insurance Institute for Highway Safety |
| Period | Through June 2026 | 2021 to 2024 |
| Miles | About 271 million rider-only miles | About 50 million driverless miles |
| What counted | Crashes with injuries, anyone at fault | Police-reportable crash involvement |
| Cyclist result | 86% fewer, 9 crashes vs about 65 expected | 3 crashes involving a cyclist, about the same rate as human drivers. Too few to say whether Waymo is safer |
| Bottom line | Promising | Not yet confirmed |
That cyclist row is not a contradiction. It measures a different outcome, over an earlier period, with about one-fifth the miles. So the cyclist benefit is not yet independently confirmed.
Only Waymo voluntarily reports its driverless miles, according to IIHS. Without miles, crash counts from other companies can’t become rates. Real riders have still been hurt, and those cases show where the trouble sits.
Where Self-Driving Cars Have Still Let Cyclists Down
The documented trouble sits at doors, curbs and close quarters, more than with a car blindly missing you on open road.
The Door That Opened Into a Bike Lane
Dooring is a city rider’s nightmare. You hold your line in the bike lane and a rear door swings into your path with nowhere to go.
Waymo says its Safe Exit system is meant to catch that moment. It gives audio and visual alerts when a cyclist is approaching as a passenger prepares to get out, while the roof display signals a stop where a door may open.
The court complaint in Hanki v. Waymo, filed in San Francisco Superior Court in June 2025, alleges that system failed. It alleges a Waymo stopped in a no-stopping zone beside the 7th Street bike lane in San Francisco, California on February 16, 2025. It also alleges a passenger opened a rear door into the lane, the rider struck the door and was thrown into a second Waymo, and the passenger received no alert.
The complaint says the rider suffered serious bodily injuries and was taken by ambulance for emergency treatment. These are allegations only.
🚪 I ride outside the door zone whenever I can, robotaxi or not. A lit roof sign on a stopped car at the curb is your cue to take more room early.
Stopping in Bike Lanes
Pick-ups are where good intentions meet bad curb space.
Christopher White, executive director of the San Francisco Bicycle Coalition, says Waymo told advocates that expecting it to respect bike lanes is “too high a bar” because customers expect to be dropped off in them. He also says Waymos pull over into bike lanes “all the time” for pick-ups and drop-offs, which is “neither legal nor safe.” In response, Waymo said its cars are designed to take the safest action available during pick-ups and drop-offs, and will often move down the street or circle the block to find a suitable spot.
From July 1, 2026, California police can issue a notice of autonomous vehicle noncompliance to the manufacturer when an officer observes an alleged traffic violation, and the maker must report it to the California DMV.
Low Light and Turns Need Extra Caution
A 2024 University of Central Florida study of California crashes involving self-driving-system vehicles found higher crash odds than human drivers at dawn and dusk and during turns. The study was not cyclist-specific, covered several companies and used data through June 2023. Still, it points at two moments riders already fear: right hooks and fading light.
⚠️ I assume no one has seen me at junctions until the car clearly slows.
No Robotaxis in Your Town? Ordinary Cars Are the Bigger Story
Most of us won’t see a robotaxi in 2026. That doesn’t make this story distant.
Waymo carried public riders in 14 US cities as of September 1, 2026, after adding Denver, Colorado, San Diego, California and Tampa, Florida, then added Las Vegas, Nevada on September 14, 2026. For most readers, the tech that matters is already parked in driveways.
That tech is automatic emergency braking, or AEB. It brakes the car when a crash looks imminent, without waiting for the driver. For riders, the question is whether it sees a bike.
In an IIHS cyclist test, a 2023 Toyota RAV4 and Subaru Forester avoided or nearly avoided a bicycle target in daylight and at night on high beams. On low beams at night, approaching a bike from behind, they only slightly slowed at test speeds of about 31 to 43 mph (50 to 70 km/h).
That low-beam failure is the surprise worth remembering, detailed in the cyclist braking study.
Bicycle AEB is neither evaluated nor required in the US, according to that 2025 study. Federal rules require pedestrian braking on new light vehicles by September 2029, with no cyclist requirement, and regulators have signaled that deadline may slip. Euro NCAP, the European new-car safety program, added a cyclist-dooring test in 2026 that rewards door-opening warnings.
Federal crash data gives the human baseline: 1,166 US bicyclists killed in 2023. With roads getting deadlier for riders, a safer-than-human robotaxi would be real progress if independent data confirms it.
How to Ride Around Self-Driving Cars (and Every Other Car)
Robots change the badge on the door. They don’t change the physics of your ride.
Give a stopped robotaxi full door-zone room. Treat a car with a lit roof sign at the curb like any double-parked car. Take the lane early rather than threading the gap.
Expect curb stops to use the bike lane. Plan the merge before the pick-up zone, signal, check over your shoulder and move out smoothly.
💡 After dark, ride as if the car behind cannot brake for you. I run lights even in daytime because being seen is cheap insurance. Pair that habit with lane position you can hold, and use these tips to be more visible when cycling after sunset.
Own the lane where doors, turns and squeezing start. Clear lane position, steady speed and early decisions make you easier for any driver or software to predict. It’s the same idea behind our guide on how to avoid dangerous encounters with drivers.
The Bottom Line
The early evidence is encouraging overall and promising for cyclists on Waymo’s own numbers, but the cyclist part is not yet independently confirmed. Documented failures center on fixable behaviors around doors and curbs, not on missing a rider in clear daylight. For most riders, the larger risk remains a human driver in a car whose braking was never required to spot a bike.
👍 My take is cautiously optimistic. I would rather pass a Waymo than a distracted human driver, but I don’t trust any car blindly until independent data catches up.
Ride as if every car is human-driven, robot or not. Hold your line, claim door-zone room, light up and never lend your safety to software you cannot see.
Know a riding buddy who needs this? Share it below👇
