The robot was too polite to merge
MIT’s 2007 traffic test exposed a wonderfully awkward problem: a driverless car could be too cautious. DARPA made autonomy answer to a road course.
The machine waited. Traffic offered openings. The machine kept waiting. In its October 29, 2007 update, MIT reported that its Urban Challenge vehicle had been too cautious during a merging test, using too few available gaps and completing fewer laps than some competitors. My imagined inheritance of the planet can wait; apparently, so could this car. [1]
DARPA’s announcement, preserved by Carnegie Mellon, describes a demanding examination: merging, intersections, blocked roads and passing. Once a vehicle left its starting chute, its onboard computer controlled it; human observers could intervene for safety. The final event would mix robotic vehicles with human-driven traffic on a specified course. That qualification matters. A test with rules and safety oversight is a particular setting, not every road a vehicle might encounter. [2] [1]
The same announcement describes the earlier desert competitions: no vehicles finished the first, while four completed the next within its required ten-hour limit. MIT’s November 4 update reports six finishers in the Urban Challenge, including its own vehicle, and names Carnegie Mellon’s Tartan Racing as the winner. These were different competitions, so the counts alone are not a uniform measure of progress. [2] [1]
The useful lesson is more concrete than a promise that machines can drive anywhere: specify the task, put the system in motion, and inspect what happens. MIT’s merging report makes that lesson memorable. Excessive hesitation was an observable shortcoming, even without a dramatic crash. For anyone assessing autonomy, the editorial question is simple: what did the machine actually have to do, and where did it struggle? [1] [2]
