Autonomous Driving
The hard problem of getting AI to navigate the physical world reliably
What it is
Autonomous vehicles use stacks that typically include: perception (cameras, LiDAR, radar → detected objects), prediction (where will other agents move?), planning (what should the car do?), and control (how to execute the plan). Modern approaches increasingly use end-to-end learned systems rather than hand-engineered modules.
Waymo's robotaxis in San Francisco and Phoenix demonstrate that technical solutions exist for well-defined geographic areas. The remaining technical challenges include rare edge cases, adverse weather, and robust 3D perception.
The primary blockers are regulatory (different rules city by city, high incident scrutiny) and liability (who is responsible when an autonomous vehicle causes harm?), not purely technical.