Why Robots Still Struggle With the Real World: Mehmet Doğar

Robots are getting smarter, but the real breakthrough is not in making them do everything. It is in teaching them to do one hard thing really well.

Mehmet Doğar, Professor of Robotics and AI at the University of Leeds, explains why object manipulation is one of the toughest problems in robotics and why the future may belong to specialized robots, not humanoid generalists.

From cluttered fridges to warehouse packing, this conversation reveals why robots still struggle when objects are hidden, partially visible, deformable, or interacting in unpredictable ways.

Key Topics:

• Why perception breaks down in cluttered environments

• Why today’s physics models are still too slow and too approximate for robot decision-making

• How robots use world models to predict what happens before they push, move, or grasp an object

• Why warehouse packing and online delivery are among the most immediate real-world applications

• What current AI hype gets wrong about robots, and where the real opportunity is