Advancing Autonomy: AI-driven Perception and Decision Making for Field Robotics

Matthew Johnson-Roberson (Vanderbilt University)

Abstract:The convergence of artificial intelligence and robotics has catalyzed transformative advancements in mobile robotics, reshaping their capabilities in real-world applications. Today's robots operate in increasingly complex and dynamic environments, requiring robust and adaptive perception systems driven by deep learning and advanced AI methodologies. This talk explores recent breakthroughs at the intersection of AI and robotics, with specific emphasis on cutting-edge research including photorealistic 3D underwater terrain generation using latent fractal diffusion models, ambiguity-free spatial foundation models through decoupling depth ambiguity. We further discuss innovative methods in teaching robots periodic stable motion generation and interpreting human sketches for robot navigation, alongside advancements in neural illumination, Gaussian splatting for trajectory generation, and photometric registration of 3D models. Insights derived from these recent contributions illustrate practical outcomes and underscore the evolving role of AI in enabling mobile robots to operate safely, reliably, and autonomously in diverse and challenging environments.

Biography: Matthew Johnson-Roberson is the inaugural Dean of the College of Connected Computing and a Professor at Vanderbilt University. Previously, he served as the Director of the Carnegie Mellon Robotics Institute and a Professor of Computer Science. Prior to that, he was a Professor of Engineering at the University of Michigan. He received his PhD from the University of Sydney in 2010, followed by postdoctoral appointments with the Centre for Autonomous Systems (CAS) at KTH Royal Institute of Technology in Stockholm and the Australian Centre for Field Robotics at the University of Sydney. He co-founded Refraction AI, a last-mile autonomous vehicle delivery company. His research career has focused extensively on robotic perception since the first DARPA Grand Challenge, emphasizing enabling robots to better see and understand their environment.