Modern perception systems are highly complex. They operate on high-dimensional data and often use black-boxed approaches such as deep learning to make sense of the world. The complexity and lack of understanding in these perception systems are tolerable in many settings, but when deploying these systems on robots with real-world consequences, it is essential to ensure expected behavior in all environments. This project will strengthen robust perception from the context of robotic manipulation. To achieve this goal, we will generate synthetic vision data using model-based and data-driven methods to adversarially test modern perception systems. Then, with a better understanding of inputs that cause perception failures, we will develop new systems that can provide guarantees for robust behavior.

I am participating in SuperUROP because I want to use skills learned from previous UROPs and classes such as Robotics: Science and Systems (6.141) to dive deeper into robotics research. I have always had an interest in robotics for the multidisciplinary skills the field requires. Through this project, I hope to contribute to cutting-edge research in robotic perception.