The goal in this project is to resolve a reference in a goal, like "the coke can on the table to the left" when the robot is operating in an open, uncertain environment. The robot might need to sense where the coke can, and table are. It might not know if a coke can even exists in the part of the environment it can access. I presented this work as a talk a workshop at RSS in 2018, and at MIT EECS Conference through a poster presentation.
This area is learning the policies for primitive motor skills. I attacked this many different ways. This started with using policy gradient descent to learn erasing. Then I learned a value function for a parameterized stirring, the writeup for which can be found at the following link:
Learning to Stir using Force-Feedback (not peer-reviewed)
I also trained preimages for pour and push, which are now used in Learning Task and Motion Planning
One of my current lines of research is learning to scoop by combining positional and force information using Dynamic Movement Primitives
Motor skill learning using sensor trace synthesis (not peer-reviewed)
From 2011-2014, I research nanoparticles that emit light. By changing the properties on their surface, I was able to control their rate of growth. I presented at Siemens, NYCSEF for 2 years, the National Junior Science and Humanities Symposium, and for the Davidson Fellowship.