My ongoing research (september 2015) is working on developing more robust walking controller for the Cornell Ranger. The key motivation behing my controller design is that the controller should be robust to errors in modelling (eg. design in simulation, directly implement on robot) as well as to external perturbations (eg. not stumble and fall when the robot encounters a hill). One of the ways that I do this is to include disturbances in the design process.
As of now, I have a working simulation that is a pretty good model for the robot, as well as a controller design process that uses some basic machine learning and optimization. I coded up the controller on the Cornell Ranger, and I can take the optimal parameters directly from simulation and get them to work on the real robot. At this point I'm working on more sophisticated controller design as well as a better characterization of the differences between the real robot and the simulation.
Conference paper for ICRA 2015. The key contribution is a way of posing and solving the following constrained optimization problem: "Find a controller that can regulate walking speed while preventing all falls, given any (bounded) disturbance."
I took my A exam in August 2014. My proposal was on the topic of 'Heirarchical Biped Control'. The idea is to do step planning on a reduced model of the robot, which is then used to help generate the motor commands on the full model. In addition to my presentation, I submitted three written reports.
For my qualifying exam (January 2012) I presented a simulation of a tractor trailer truck driving backwards along a curving road. It had partial sensing of its state using noisy sensors, and control of the steering angle and travel speed (both of which were bounded). State estimation was done using an extended Kalman filter (EKF). Control was done by using a discrete linear quadratic regulator (dLQR), on the linearized system (about the current state estimate).
My undergraduate thesis project at Tufts University was designing, fabricating, and testing a new type of surgical bone saw. At the conclusion of the project I submitted provisional patent for my saw design and published the experimental results in the ASME Journal of Medical Devices. I was also worked on two conference papers - one for my saw, and one for a saw designed by another student.
Research advisor: Tom James
In my last year of undergrad, I worked as a research assistant in the Non-Newtonian Fluids lab at MIT, along with Tim Lannin and Will Langford. Our goal was to design and implement a non-linear tracking controller on their Filament Stretching Extensional Rheometer (FiSER) and to redesign the LabVIEW program that ran the experiments.
FiSER characterizes fluids by rapidly stretching them, and measuring both the axial force on the fluid and the diameter of the sample in real time. The original experiments were done by using open-loop velocity profiles for the two measurement stages of the device. We implemented a controller that adjusted the speed of the stages in real-time to track a desired diameter profile in the specimen. This was challenging because the reference trajectories were all exponential, requiring the use of non-linear control techniques. The data acquisition, control, and estimation code were written in LabVIEW and compiled to a National Instruments cRIO and FPGA.
Research advisors: Chris Rogers and Gareth McKinley.
As an freshman at Tufts University, I took a chemistry course, which was designed to teach chemistry by attempting to answer the question: "Is hydrogen the fuel of the future?". My final report for the class was included in supplement of the paper that my professor wrote about the course.