New challenge for AI

A team at the University of Tübingen created a new system called Fail2Drive that throws unexpected obstacles at autonomous cars, like an elephant in a crosswalk, a playground slide in the road, or a wall painted to look like open pavement. The cars failed all of it — crashing into obstacles and mistaking fake roads for real ones.

Overall, success rates dropped by nearly 23 percent, showing that many self-driving systems may be memorizing patterns, not truly adapting to the unexpected.