Certifiable Outlier-Robust Geometric Perception - Robots that See through the Clutter with Confidence

Explore the cutting-edge field of certifiable outlier-robust geometric perception in this one-hour lecture by Heng Yang from MIT LIDS. Delve into the challenges of estimating geometric models from sensor measurements in robotics applications, and learn about innovative algorithms designed to provide performance guarantees despite the presence of outliers. Discover how graph theory and graduated non-convexity are employed to prune gross outliers and compute optimal model estimates, and understand the role of sparse semidefinite programming in providing optimality certificates. Examine practical applications in robotics, including scan matching, satellite pose estimation, and vehicle pose and shape estimation. Gain insights into the integration of certifiable perception with big data, machine learning, and safe control for trustworthy autonomy. This Paul G. Allen School presentation offers a comprehensive look at advanced techniques for enhancing robot perception and decision-making in cluttered environments.