IST LUNCH BUNCH

Tuesday February 6, 2018 12:00 PM

Nonlinear control and estimation theory for multi-agent autonomy and autonomous flying ambulances

Speaker: Soon-Jo Chung, Engineering and Applied Science, CALTECH
Location: Annenberg 105

Recent advances in self-driving car and drone technologies are turning a century-old dream of vertical-take-off-landing personal transportation vehicles into a reality with many existing projects in development. Caltech's Center for Autonomous Systems and Technologies (CAST)'s engineers and scientists have developed a 1/5 working scale model of their Autonomously Flying Ambulance (AFA) with innovative design ideas, including flight by a hybrid of distributed fans and deployable wings, bio-inspired flight and control, and vision-based navigation. The model has been flight-tested successfully in CAST's unique drone arena using an open-air distributed fan-array wind tunnel. CAST's AFA rotorcraft and autonomy technologies can provide solutions for a range of short-distance travel challenges: point-to-point delivery of packages on Earth or scientific samples on Mars. I will review some of the control theoretical results derived for control and coordination of novel aerial robotic platforms.  First, I will present distributed, motion planning and multi-point routing algorithms for optimally reconfiguring swarms of vehicles with limited communication and computation capabilities from various pick-up locations to target locations. The real-time guidance algorithm solves both the optimal assignment and collision-free trajectory generation in an integrated manner. Three related approaches have been derived for optimal assignment problem for real-time routing: (1) distributed auction assignment, (2) novel probabilistic swarm guidance that employs time-inhomogeneous Markov chains; and (3) potential games solved by binary log-linear learning. Second, nonlinear tracking control and estimation is utilized to track optimal reconfiguration trajectories with a property of robustness (finite-gain Lp incremental stability). I will also show such nonlinear incremental stability analysis can be extended to a set of Itô stochastic nonlinear systems for synchronization control and nonlinear estimation, including exponential stability of a distributed Bayesian filtering algorithm, robust nonlinear estimation for visual SLAM, and consensus stability of distributed reinforcement learning for flying ambulances or taxis.

Series IST Lunch Bunch

Contact: Diane Goodfellow diane@cms.caltech.edu