"Engineering Emergence in Robotic Systems"

Abstract:
Complex systems consist of diverse agents that are interdependent, connected, and continually adapt to changes in the system and environment. The defining feature of complex systems is the phenomenon of emergence, which describes the spontaneous creation of order and functionality from the bottom up. We see emergent patterns at every level in the physical world, both in the structure and functionality of complex systems. Some emergent phenomena are desirable, for example, the swirling patterns created by flocks of starlings or the natural formation of lanes in heavy pedestrian traffic. Other emergent phenomena are undesirable, such as phantom traffic jams that emerge from local interactions between vehicles on a highway. Emergence enables a system to accomplish tasks that are unachievable by any individual agent while remaining robust to noise, disturbances, and agent failure. This makes an emergence framework ideal for multi-robot and swarm applications, where monetary constraints significantly restrict the sensing, actuating, and communicating ability. In this talk, I will present an approach to exploit the mechanism of emergence through the rigorous application of decentralized optimal control. I will start with a multi-level iterative control framework, which is explored through a case study on the desirable emergence of highway platooning with connected and automated vehicles. This leads to intuitive interpretations of the sufficient conditions for the emergence, and suppression, of platooning behavior. In the second part of the talk, I will rigorously justify the use of integrator dynamics for planning in robotic systems via differential flatness. I will conclude the talk with a vision of future research and several fundamental open questions.