Industries / Transport and vehicles / Ride-share and taxis
Driverless cars
This is the work of carrying paying passengers in cars that drive themselves, with no one at the wheel.
6
Milestones achieved, from 2018 to 2026.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
2018
A test car kills a pedestrian
On March 18, 2018, an Uber test SUV driving itself in Tempe, Arizona, struck and killed a woman walking a bicycle across the road. Its software detected her 5.6 seconds before impact but never correctly classified her, and emergency braking was not designed to act. The National Transportation Safety Board (NTSB) found the safety operator had been looking at her phone, and blamed Uber's "inadequate safety culture" as a contributing factor.
LessonA human told to watch a machine that rarely fails soon stops watching, so a safety plan cannot rest on that person alone.
Credit
NTSB (report HWY18MH010)
2020
The public rides with no driver
On October 8, 2020, Waymo opened rides with no one in the driver's seat to the public in metro Phoenix, starting with existing Waymo One riders. Before that, only 5–10% of its rides were fully driverless, all for riders bound by non-disclosure agreements.
Credit
Waymo
2023
A permit pulled after a crash
On October 2, 2023, a human-driven car in San Francisco threw a pedestrian into the path of a Cruise robotaxi, which then tried to pull over and dragged her about 20 feet (6 m). On October 24, 2023, California's Department of Motor Vehicles (DMV) suspended Cruise's driverless permits, citing an "unreasonable risk to public safety," and said Cruise had not first shown it the full video; Cruise disputed that. The state Public Utilities Commission also suspended its passenger service.
LessonWhat a company tells regulators after a crash can matter as much to its permit as the crash itself.
Credit
California DMV; California Public Utilities Commission
2024
An insurer counts the claims
Swiss Re, the reinsurer, and Waymo compared insurance claims over 25.3 million driverless miles in Phoenix, San Francisco, Los Angeles and Austin with a human baseline of over 200 billion miles. They found 9 property-damage and 2 bodily-injury claims where human drivers would be expected to produce 78 and 26, an 88% and 92% reduction. The study was written with Waymo and had been submitted to a journal.
Credit
Swiss Re; Waymo
2024
GM stops funding Cruise
On December 10, 2024, General Motors said it would stop funding Cruise's robotaxi work and focus on driver assistance for cars it sells, citing the cost of scaling and the competition. GM had put more than $10 billion into Cruise since 2016, Fortune reported.
LessonHaving money and a car factory was not enough without public trust and time to scale.
Credit
General Motors
2026
An outside test of crash rates
The Insurance Institute for Highway Safety (IIHS) compared about 50 million Waymo driverless miles with about 222 billion human miles in the same four cities. Counting only crashes people would usually report to police, it found 68% fewer crashes per mile overall, 76% fewer in Phoenix and 35% fewer in San Francisco; Austin was 4% higher on a small sample. IIHS could not rate other companies because only Waymo publishes its mileage.
Credit
IIHS, July 23, 2026
Watch for
A million paid rides a week
Waymo co-CEO Tekedra Mawakana said Waymo would give "more than 1 million paid weekly robotaxi rides in the U.S. by the end of this year," up from about 400,000 a week across six cities.
Achieved whenAlphabet's filings or an outside count confirm more than 1 million paid rides in a week.
Claimed by
Waymo, February 2026
Tesla rides with no one aboard
Tesla began robotaxi rides in Austin with no safety monitor in the car on January 22, 2026, starting with "a few unsupervised vehicles" and with what appeared to be chase cars following them. In June 2026 Electrek, citing a volunteer-run tracker, counted 14 of 31 active Tesla robotaxis running without a monitor, against 42 vehicles Texas had authorized.
Achieved whenTesla publishes driverless mileage and an outside body such as IIHS or a state regulator reports routine unsupervised service.
Claimed by
Tesla, January 2026
Open
01
A clean federal crash file
A cleaned copy of NHTSA's crash reports for driverless cars, updated monthly and published with its rules. NHTSA is the US road-safety regulator, and its Standing General Order requires companies to report these crashes. Remove duplicates and off-road events, flag likely police-reportable crashes, and pair each company with the best available mileage estimate.
The briefCredit · unclaimed02
Read the crash stories
A labeled test set of crash narratives from the federal file, each marked by two people for who struck whom, whether anyone was hurt, and whether the car was driving itself, then a score of how well an AI model applies the same labels.
The briefCredit · unclaimed
Open 01 · the brief
A clean federal crash file
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A cleaned copy of NHTSA's crash reports for driverless cars, updated monthly and published with its rules. NHTSA is the US road-safety regulator, and its Standing General Order requires companies to report these crashes. Remove duplicates and off-road events, flag likely police-reportable crashes, and pair each company with the best available mileage estimate.
Why it is still open
IIHS had to discard about a quarter of the records and judged only 22% of the remaining 736 crashes likely police-reportable. Its president said "the present data collection system isn't good enough to allow continuous monitoring of a large-scale expansion." Only Waymo reports miles.
Built with what exists
NHTSA Standing General Order public data files; Python and DuckDB; company mileage disclosures and California DMV and Texas permit records for fleet size; a public code repository.
First step
Download the SGO files and try to rebuild IIHS's count of 736 cleaned crashes, writing down each rule you have to guess.
Done looks like
Three monthly releases in a row, with duplicate rules published, that reproduce the IIHS counts for Waymo within 5% for the period IIHS studied.
Open 02 · the brief
Read the crash stories
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A labeled test set of crash narratives from the federal file, each marked by two people for who struck whom, whether anyone was hurt, and whether the car was driving itself, then a score of how well an AI model applies the same labels.
Why it is still open
IIHS had to sort the reports by hand to decide which crashes counted. Electrek reported that Tesla had redacted its crash narratives before releasing them. This research found no public labeled set of these narratives.
Built with what exists
NHTSA SGO narrative fields; an open labeling tool such as Label Studio; an AI model that reads long text (the site's note "Seven more flagged" covers one such model checking documents against people).
First step
Pull 100 narratives from one quarter, label them yourself, and write the label rules down where you hesitated.
Done looks like
At least 1,000 narratives labeled by two people with their agreement rate published, and the AI model's agreement with them reported per label.