Title: Leveraging Computer Vision and Reinforcement Learning for contact and non-contact based plant phenotyping.  

Abstract: 

Effective plant breeding requires scientists to find correspondences between genetic markers and desirable physical traits (phenotypes) of the genotype. Robotics can aid the acceleration of breeding pipeline by facilitating high throughput plant phenotyping. In this thesis we propose a computer vision pipeline for non-contact based plant phenotyping and a Reinforcement Learning framework to learn visiomotor policies for contact based phenotyping.  Reinforcement learning is known to be sample inefficient, preventing its application to many real-world problems, especially with high dimensional observations like images. Learning auxiliary tasks along with the reinforcement learning objective could be a powerful tool to improve the learning efficiency. We propose a principled online learning algorithm that dynamically combines different auxiliary tasks to speed up training for reinforcement learning.

Committee:

George Kantor (advisor, RI)

David Held (RI)

Brian Okorn (PhD, RI)