In the competition, a Fetch mobile manipulator robot autonomously navigates to stations in a narrow work cell to pick parts in irregular shape (screws, gears, gearbox parts) and machine large gears by inserting into a narrow chuck, transport and place them into a human-friendly caddy, and drop off the caddy on a specified region at inspection station.
The contest includes the full range activities commonly found in manufacturing environments. As the media reported, “While conceptually simple, combining autonomous mobility with robotic arm activity is extremely challenging, requiring complex interaction between the robot navigation, machine vision, arm operation and safety systems.”
Our navigation code enables the Fetch robot to autonomously navigate in the narrow space and stop closely to tables for manipulation; compared to popular hard-coded solutions with the robot not aware of table obstacles. Our perception work is able to successfully and rapidly detect screws, small gears, gearbox parts and the almost symmetric caddies, which are all of irregular shape not commonly seen in academic work. Our manipulation work enabled safety, being able to avoid not only common obstacles – walls and tables – but also complex ones, e.g., the machine with door and a chuck inside.