Seed breeding is a critical part of agriculture and ensures food security through the development of new crop varieties that are higher-yielding, disease and drought resistant and adapted for each region. Best practise is to have breeders measure and record traits across a large number of seed trials. However, it is hard to get consistent measurements across breeders especially across different regions. There are also limits to what a person can see and record. It also generates a large amount of data that needs to be manually entered for analysis.
The Deepfield robotics team developed a system for Automated Field Trials with the tag line “From the internet of fields to the internet of plants.” They designed a 4D-Scan robot and Deep Cloud analysis tool. The 4D-Scan robot could record detailed 2D and 3D imagery of plants as well as their GPS location. Each plant could be tracked across multiple days by their GPS location. The Deep Cloud analysis tool could count plants including recording the exact emergence and mortality, measure leaf area and height, and correlate these with temperature and time.
I helped at the tail end of this project until it was paused while we investigate other topics.
This video has a good introduction to the project:
More technical details are in a talk I gave for the team at ROSCon 2016 in Seoul, Korea.