Verifiable Learning and Control of Multi-Robot Systems On The Fly

Multi-robot systems play a vital role in complex autonomous missions, offering unique advantages over single-robot setups.

Such missions often require complex robotic structures, such as robotic manipulators, to deliver tasks with deadlines in obstacle-cluttered and highly dynamic environments that induce uncertainties, disturbances, and abrupt, unexpected events.

Existing works fail to address these challenges since they either consider simple environments and robots with unrealistically simple and known dynamics or adopt centralized approaches that cannot scale to large teams of robots.

The purpose of this project is to develop theory and algorithms for safe and scalable learning-based coordination of multi-robot systems for the accomplishment of complex time-constrained tasks.

We will achieve this by creating novel methodologies at the intersection of control theory, learning, robot motion planning, and computer science.

The resulting algorithms will (i) be scalable for real-time operation by large teams of robots, (ii) establish verifiable guarantees concerning robot safety and the execution of time-constrained multi-robot tasks, and (iii) learn and adapt on the fly to uncertainties, disturbances, and unforeseen events.

The research outcomes have the potential to pave the road for the development of truly autonomous systems and impact industries such as healthcare, manufacturing, and disaster scenarios.