Bimanual robots often need to move both arms in conjunction. Planning for such motion deals with non-convex landscapes because of the nonlinear equality constraints between the end
effectors. Trajectory optimizers are finicky with such constraints, needing good initializations for feasible solutions. Also, the valid configuration space is a measure-zero set, forcing sampling based methods to rely on approximations or projections. A newer approach decomposes the planning space into convex sets. First we decide on a higher level the sequence of sets to move through. Then we find trajectories through each set, using convex optimization to guarantee a collision free path if it exists. Working off this approach, my research will start with investigating approaches for optimizing over more complex objectives while maintaining feasibility guarantees. Ultimately, I aim to improve how we plan and execute motion in bimanual robots.

I am very excited about working with the Robot Locomotion Group on research in a field of robotics I have only really taken (and enjoyed!) classes in before. Throughout the year, I hope to hone essential research and communication skills in the structured environment of SuperUROP and make publishable progress on my project. I am also looking to get a more structured introduction to what doing research means and how to approach it as I consider whether I want to work towards going to graduate school in the future.