Radim Tylecek presents Challenges in autonomous outdoor gardening with a robot using passive vision

On 2019-11-11 - 2019-11-11 14:30:00 at G205, Karlovo náměstí 13, Praha 2
The TrimBot2020 project has developed a vision-controlled robot for
autonomous navigation and trimming of gardens, which works in uncontrolled
outdoor conditions. The robot navigates over varying terrain and
approaches rose bushes
and boxwood topiary, to trim them to an ideal shape. The platform is derived
from the Bosch Indego robot lawnmower and is equipped with 5 pairs of
stereo cameras that provide a 360 deg view. It navigates using a
user-defined garden map and 3D scene analysis and then visually servos
to align itself to its trimming position next to the bush.
Trimming of the bush is done using a novel electric plant cutter
attached to the Kinova Jaco arm, which is mounted on the platform.
Trimming occurs after scanning of the bush shape from an arm-mounted
stereo camera.
Two different actuators and control algorithms were designed for topiary bushes
(rotating blades) and rose bushes (stem clipper).
Achieving this garden application required a combination of robotics
and 3D computer
vision research and innovation activities. Original developments in
3D sensing of semi-regular surfaces with physical texture
(overgrown plant surfaces), coping with outdoor lighting variations,
identifying different objects and types of surfaces,
self-localising and navigating around obstacles, visual servoing to
potentially moving target plants, leaves and branches, and innovative
engineering were required to deliver all this on a small battery-powered
consumer-grade vehicle.
Responsible person: Petr Pošík