The kidnapped robot problem is a benchmark test for a robot's localization system. It occurs when a robot is physically picked up and moved to a different part of its environment without its odometry sensors registering the movement. To solve it, the robot must realize its internal map no longer matches its sensor data and successfully re-localize itself from scratch.
Heard on the job
The new firmware handles dynamic obstacles well, but it completely fails the kidnapped robot problem if an operator manually pushes it across the warehouse.
Insider tip
A robust particle filter is the standard approach to solving this, allowing the robot to maintain multiple hypotheses of its location until the sensor data converges.