Industries / Government, education and community / Education / Driving schools
Classroom teaching
Classroom teaching covers the rules of the road, risk and hazard awareness, and the knowledge a learner needs before and alongside lessons in the car, taught in schools, in driving-school classrooms or online.
6
Milestones achieved, from 1933 to 2016.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1933
A school driving course
Amos Neyhart, a professor at Penn State, began teaching volunteer students from State College High School in Pennsylvania to drive in 1931, using his own 1929 Graham-Paige car. His first formal courses began in 1933, and he later gave the first course that trained teachers to teach driving. By 1968 he had trained almost 20,000 driving instructors, by the count of Ford's traffic safety department. Sources give different start years: Education Week's obituary says 1931 and 1933, a New York Times obituary quoted by Onward State says 1932, and the Encyclopedia of Education says 1934.
Credit
Amos E. Neyhart, Pennsylvania State University, with support from the American Automobile Association (AAA)
1980
Courses cut, crashes fall
A 1978 study of 27 US states found that high school driver education led many 16- and 17-year-olds to get a license earlier than they otherwise would have. It did not lower fatal crashes per licensed driver. The authors estimated that about 80% of 16- and 17-year-olds who took the course got licenses they would not otherwise have had until age 18 or later. In 1976 Connecticut stopped funding the course, and nine school systems dropped it. A 1980 follow-up by the same lead author found that in those communities fewer 16- and 17-year-olds got licensed, and crashes in that age group fell substantially.
LessonA course that gets teenagers licensed sooner can add crashes, because it puts more young drivers on the road, even if each driver is no worse.
Credit
Leon S. Robertson and Paul L. Zador
1986
Bigger course, more crashes
In the DeKalb County, Georgia, study, funded by the National Highway Traffic Safety Administration (NHTSA), students were assigned to an enhanced course (the Safe Performance Curriculum), a minimal course, or no school course. It was the largest experimental test of high school driver education. A 1986 reanalysis found that students given the enhanced course were more likely than the control group to get licensed, to crash and to get traffic violations. Sources disagree: the Encyclopedia of Education says the full course cut crashes for six months before the effect faded.
LessonTraining more skills did not make students safer, and the longer course led to more driving earlier.
Credit
NHTSA and DeKalb County schools; reanalysis by Adrian Lund, Allan Williams and Paul Zador
2001
Review finds no crash benefit
A Cochrane systematic review pooled three randomized trials of school driver education, run between 1982 and 1984 in the US, New Zealand and Australia with 17,965 students. Students who took the courses got licensed sooner. In New Zealand, males took 111 days to get licensed after the course and 300 days without it. The trials showed no evidence that the courses reduced crashes: 27.5% of course students crashed in the US trial, against 26.7% of controls.
Credit
Ian Roberts and Irene Kwan, Cochrane Injuries Group
2014
Modest gains in Oregon
In September 2014 the AAA Foundation for Traffic Safety published a study of beginner driver education in Oregon and Manitoba. It compared the driving records of 94,342 Oregon teens and found 4.3% fewer collisions and 39.3% fewer traffic convictions among those who took the course. In both places, students who took the course passed the road test at about the same rate as those who did not, and in Manitoba they still failed to spot many hazards. The study called the effects "positive but modest."
Credit
AAA Foundation for Traffic Safety
2016
Hazard training tested on crashes
RAPT (Risk Awareness and Perception Training) is a PC program of about an hour that teaches new drivers where to look for hazards they cannot yet see. In an on-road test, eye tracking showed trained drivers looking for the hidden hazard 65.8% of the time, against 47.3% for an untrained group. Six or more months later the figures were 61.9% and 37.7%. NHTSA's 2016 evaluation of crash records of newly licensed teens in California found no effect overall. Trained males had a 24% lower crash rate than untrained males, and females showed no significant difference.
LessonTraining that changes where learners look can show up in crash records, but not for every group.
Credit
Donald Fisher, Alexander Pollatsek, Anuj Pradhan and colleagues at the University of Massachusetts Amherst; NHTSA
Watch for
Theory without the classroom
Germany's government proposes ending required attendance at driving-school theory lessons. Learners could learn the material entirely by a digital route such as an app, and each school would decide whether and how to offer theory lessons. The ministry puts the average class B car license at about €3,400 and says the reform should lower it. The Bundestag passed the first part of the reform in November 2026. The Bundesrat, the chamber of the states, was due to take it up on 20 November 2026. The start date is "early 2027" by the ministry's May 2026 account and July 2027 by Mobiflip's November 2026 report.
Achieved whenThe rules are in force and the first learners who studied theory only by app have sat the theory test, with results published.
Claimed by
German Federal Ministry for Transport (BMV), May 2026
AI-written revision tests
A web platform for revising for the UK driving theory test uses generative AI to write new question sets and to give feedback that adapts to each learner's past results. Its authors say the generated questions closely match tests written by experts. The platform was judged by expert review and model analysis, not by learners' results.
Achieved whenA study with real learners shows they score higher on a theory test after using it than learners using standard material.
Claimed by
Riya Gill, Ievgeniia Kuzminykh, Maher Salem and Bogdan Ghita, September 2025
Open
01
Measure the move online
A public before-and-after study of Germany's theory test results around the date when classroom attendance stops being required, broken down by federal state. Add a survey of how candidates studied: by app only, in a classroom, or both.
The briefCredit · unclaimed02
Trial an AI theory tutor
A randomized study, with consent from every participant, in which learner drivers revise for a theory test either with an AI tutor or with the official study material, and then all sit the same mock test. Also count how often the tutor's explanations contradict the official handbook.
The briefCredit · unclaimed
Open 01 · the brief
Measure the move online
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public before-and-after study of Germany's theory test results around the date when classroom attendance stops being required, broken down by federal state. Add a survey of how candidates studied: by app only, in a classroom, or both.
Why it is still open
Germany's reform will let learners study theory entirely by app, and the rules are planned to start in 2027. The Federal Motor Transport Authority (KBA) counted 2,042,050 theory tests in 2025, of which 828,324 were failed. Its tables count tests, not people, and do not record how candidates studied. A plan in the reform to publish each school's pass rate has been set aside for now, by Mobiflip's account of criticism from the ACE motoring club.
Built with what exists
KBA's yearly FE 7 test spreadsheets, Python with pandas, a free survey tool such as LimeSurvey, and a preregistration on the Open Science Framework.
First step
Download the FE 7 tables from 2017 to 2025 and publish the yearly theory failure rate for each state as a baseline.
Done looks like
Theory failure rates for all 16 states for at least three years before and one year after the change. A survey of at least 1,000 candidates linking how they studied to their first-attempt result. Code and data published under an open license.
Open 02 · the brief
Trial an AI theory tutor
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A randomized study, with consent from every participant, in which learner drivers revise for a theory test either with an AI tutor or with the official study material, and then all sit the same mock test. Also count how often the tutor's explanations contradict the official handbook.
Why it is still open
Published work tests the AI models, not the learners. IDKB (September 2024) scored 15 vision-language models on over one million items from several countries. DriveQA (August 2025) tested models on traffic rules and right of way. A September 2025 platform for UK theory revision was judged only by expert review. Searches in November 2026 found no randomized trial measuring learners taught by an AI tutor, while Germany plans to let learners study theory by app alone.
Built with what exists
A current large language model through its API; one jurisdiction's public driver handbook and the licensing authority's sample questions; a quiz platform such as Moodle; R or Python for the analysis.
First step
Write and preregister the protocol, and build a 100-question mock test from one jurisdiction's public handbook.
Done looks like
At least 200 learners randomly assigned, an analysis plan registered in advance, mock test scores for both groups, and a count of tutor errors per 100 answers, all published.