- Research Program Mentor
PhD candidate at University of Pennsylvania (UPenn)
Robotics, Mechanical Engineering, Mechanical Design, Product Development
BioI am a PhD candidate at the University of Pennsylvania studying how we can use social robotics and computer vision to improve access to rehabilitation over telepresence. I am interested in how robots can interact with and measure people. I previously focused on designing and manufacturing medical devices and spent a while working on computer vision and collaborative robots for manufacturing.
Designing a Robot to Elicit Emotional Responses
In this project you will design a robot from the neck up to convey emotions to people. Robotics is growing rapidly, but most people don't really know how to understand them. You will begin by surveying the state of the art in robotic design, looking at both research papers and robotics out in the wild. With the help of methods proposed in the literature, you will develop an understanding of what components would allow a robot to convey its emotions. Perhaps you will find that a screen with faces displayed and a neck that can look side to side is best. Maybe eyebrows are really important. Your reading will guide you. Once you understand what is important, you will sketch out a robot head and neck. You will use cardstock, bristol board, hobby servos, screens, arduino, etc. to put together a prototype. You will then program your new robot to express a few different emotions. You will test your robot's ability to convey emotions with a few friends and family. Finally you will report on your findings in a short paper, presenting the literature that you found, your approach to design, test methods, final design, results, and directions for future work.
Understanding Human Motion from Video
In this project you will apply off the shelf machine learning based systems to track people moving and say something about the quality of their motion. As people age, recover from injury, and develop new skills, the way they move changes. Understanding these changes can allow tracking of progress. You will identify a motion/series of motions that you are particularly interested in, you will record yourself and a few friends doing that motion both normally and with a simulated injury (by adding weight to your arm). You will briefly read up on a set of pose tracking algorithms and pick one to use, which will give you joint positions over time. You will then turn to the literature to identify a few key measurements that you can make that will indicate quality of motion (ex: max speed of motion, smoothness of motion). You will implement these features to calculate them across your dataset. You will then segment your data into both a training and test set. You will choose a simple machine learning classification algorithm to classify motions into normal or impaired. You will present your findings in a short report outlining the state of the art, your goals, approach, results, and how you could extend the project further.