I will discuss several recent research results on developing Super Autonomous or USARC: Unmanned, Small, Agile, Resilient, and Collaborative robots. Achieving this goal requires reimagining the existing perception-action paradigm governing robot autonomy. This involves shifting the existing paradigm from a sequential to a concurrent approach by combining in a principled manner physics-based and data-driven techniques across modeling, perception, learning, and control. This will result in an architecture which proactively and continuously improves its navigation performances and that naturally scales to multi-robot settings by leveraging data from multiple agents. Consequently, this will boost agility, resilience, and improve both individual and collaborative decentralized decision-making processes for small-scale robots.
