Robotics, Controls and Intelligence
Robotics to me is the great amalgamation of knowledge that enables people and algorithms to exert their will on the physical world.
I am interested in many applications of intelligence and controls. I am actively pursuing research in intelligent task and motion planning, dynamic modelling and control of complex-mobility robotics, with some work in robot-terrain interaction. In practice, all these components are necessary when attempting to design the intelligence stack that is deployed on a field robot.
Intelligent Task and Motion Planning
How far ahead should my robot consider the world it sees? When can I predict the motion of a moving agent? When do I know if I should recompute a new plan? (Hopefully before it is invalidated.) Can a human operator help improve motion planning decisions? Can I intelligently select way-points to get close-to-optimal plans in real-time? (MIT collaborators are very interested in this.)
I am well poised to solve these problems, each of which can be a serious research proposal and paper.
A large portion of my research energies (particularly with the US Army Research Lab) is in development of novel algorithms for motion and task planning. A robot after all is only interesting if it can perform the task that it was designed to do, and that generally requires movement.
My approach to motion planning uses model-predictive control combined with a sampling-based A* algorithm. This gives us the robustness of a model-predictive controller with the speed of a heuristics-based approach. It also generates a globally optimal trajectory, a notable feature. This motion planner is powerful and fast, letting it not only consider complex models, but it can learn and adapt online in real-time through its deep-learning networks.
There are many interesting directions that this research can take. Some of my collaborators at JPL/CalTech and I are interested in applying this planning algorithm on the newest generation legged platform (LLAMA) for the US Army.