Autonomous vehicles operating in complex environments require robust planning capabilities. Current approaches often treat mapping, localization, and planning as separate modules, leading to suboptimal behaviors when uncertainties compound. Perception-aware planning is an emerging paradigm in robotics that couples planning with perception capabilities, enabling systems to adapt to the quality of their sensory inputs. This project proposes a framework leveraging generative models to jointly predict environments, localization and trajectories. The ability to sample from a learned distribution will be exploited for uncertainty quantification across these coupled outputs. We aim to develop a perception-aware planning system that uses this structured uncertainty to make more informed decisions.

I am participating in SuperUROP to gain deeper research experience, take ownership of an extended project, and expand my knowledge in robotics and machine learning. Building on past UROPs and coursework, I am eager to explore both theory and real-world deployment. What excites me most is the chance to contribute meaningfully to the field while growing as an independent researcher.