Perception models for selective harvesting robots in fruit and vegetable production

Summary

Selective harvesting of fruit and vegetables is a task that is currently suffering from a labour shortage. To prevent that this labour shortage leads to a reduced supply of fresh fruits and vegetables, robotic alternatives are currently being researched. For a robot to be able to selectively harvest a crop, the fruits and vegetables must be perceived using sensors and perception models. Unfortunately, to date, most perception models are still unable to perform generically when deployed in different growing conditions. This is unfavourable for the commercial success rate of these robots. This PhD thesis aimed to improve the commercial success rate of selective harvesting robots by investigating localisation models and deep learning networks that can provide better performance in different growing conditions. Application of these techniques to a prototype harvesting robot has led to improved performance in several orchards and vegetable fields.