29 October 2019
The Trimbot – advertised on the project site as ‘cutting hedge technology’ - consists of an automated lawnmower with an added flexible robotic arm and five cameras. This allows the robot to navigate through a garden and to capture the structure of bushes in 3D. By comparing this structure with 3D information from well-trimmed shrubs, the Trimbot knows where to cut the branches. In addition to navigating and recognizing the 3D structure of bushes, an important challenge for the researchers was to ensure that the robot can do its work under different light and weather conditions.
The project was supervised by the School of Informatics at the University of Edingburgh. Also several Dutch scientists were involved. Theo Gevers, professor of Computer Vision at the University of Amsterdam, was part of the project leading team. The other UvA scientists involved are PhD students Hoang-An Le and Anil S. Baslamisli and postdoctoral researchers Fares Alnajar and Sezer Karaoglu.
The contribution from the UvA scientists consisted of creating a dataset with virtual garden objects, and using this to train the robot to recognize garden objects in a realistic setting and under changing circumstances. The team managed to get the robot to recognize objects correctly 92% of the time.