While robots have become increasingly important in a variety of settings, robotic manipulation planning has remained relatively limited, often limited to pick-and-place operations. However in reality, humans interact with objects through many different modes of interaction, such as pushing, grasping, or tool usage. The goal of this project is to build an hierarchical, multimodal manipulation planner that first plans in object space before refining the plan in robot space, avoiding expensive computations in the robot space that are infeasible due to object space constraints. It will further allow the robot to use different modes of interaction to manipulate objects, better approximating the variety of different modes humans manipulate objects and allowing robots to efficiently interact with objects in diverse settings.

I have worked on robotics in the Learning and Intelligent Systems (LIS) group on the PR2 humanoid robot and previously worked in the Personal Robots Group in the Media Lab. Over the summers, I’ve interned as a software engineer at VMware and Twitter and have a strong interest in algorithms, statistics, and machine learning.