Embracing
Contacts for Dexterous Manipulation
The last decade
has seen remarkable progress in the fields of deep learning, artificial
intelligence (AI) and reinforcement learning (RL). However, similar progress
has largely eluded feedback control of physical robots -- robots still operate
in highly structured and engineered environments that are designed so that
robots can avoid contact with their environment. Contacts play a central role
in manipulation. Planning and feedback control in the presence of contacts
remain very challenging problems, making design of closed-loop manipulation
systems elusive. In this talk, I will present several robotic tasks with
different degrees of complexity. Each of these systems present unique
challenges with regards to modeling, learning, sensing and optimization. I will
highlight how we address these challenges in order to achieve efficient and
generalizable performance for these robotic systems. The techniques proposed in
these problems could be instrumental in creating next-generation robotic
systems that can (possibly) plan in simulation and operate in the real world
with real-time sensing and perception. Based on these results, I will present a
vision toward creating next-generation robotic systems which can perceive and
interact with their environment with higher degrees of autonomy. These research
problems will enable a principled way to feedback robotic manipulation which
will find applications in factory automation, manufacturing, home assistance
and human-robot collaborative environments.
Bio : Devesh K. Jha is currently a Principal Research
Scientist at Mitsubishi Electric Research Laboratories (MERL) in Cambridge, MA,
USA. At MERL, he has been working on fundamental problems in the areas of robot
learning and manipulation, with applications to factory automation, e-commerce
and manufacturing. He received PhD in Mechanical Engineering from Penn
State in December 2016. He also received M.S. degrees in Mechanical Engineering
and Mathematics from Penn State. His research interests are in the areas of
Robotics, Machine Learning and Deep Learning. He is a recipient of several best
paper awards including the Kalman Best Paper Award 2019 from the American
Society of Mechanical Engineers (ASME), Dynamic Systems and Control Division
(DSCD). He is a senior member of IEEE and an associate editor for IEEE Robotics
and Automation Letters (RA-L).