The purpose of this blog post is three-fold: to chat about the major findings (including some interpretations too uncertain for the paper), to share the human side of the story, because our brains work in narratives, and to highlight the most important lessons I learned about doing useful RL research.
Let’s go!
This research kicked off my journey into Reinforcement Learning (RL) as a first-year PhD student. My background was all over the place: assistive robotics (DLR), semantic perception (JPL), and even astrophysics. I had zero prior experience in RL, but I thought it was the most compelling form of robot learning, so I was determined to study it.
I was immediately drawn to learning from tactile feedback because of the ultimate goal I had in mind: physical human-robot interaction. My interest in robotics stems from the belief that they can dramatically improve the quality of life for many people as physical assistants (less so to do my laundry).