Glen Berseth, Core Academic Member of Mila, was awarded funding by the Canada Foundation for Innovation to purchase robotics hardware, which will allow him to further his research on embodied machine learning.
Glen Berseth, also Assistant Professor at Université de Montréal, will use this grant to work on improving generalization across robotics and solving real-world learning challenges.
The new hardware will include robotic arms of various sizes, robot dogs, and platforms that can be used to train machine learning algorithms to perform humanlike manipulation tasks in industrial processes such as manufacturing, cooking food, and managing warehouses.
Another potential use is for recycling via the assembly and disassembly of electronics which will also spare humans the risks of interacting with harsh chemicals and dangerous materials.
“The idea is to increase the amount of data that we can collect to be able to train these arms,” Glen Berseth said.
“We can use the smaller and cheaper robots to collect a lot of data -which is usually a bit noisy and inaccurate- and then use a small setmore industrial arms to augment the cheap data set and eliminate much of the uncertainty by having low cost and highly accurate robots work together.”
“Then, we can train one policy that will work across all of these different robot arms for performing different tasks,” he explained.