Control of Distributed, Autonomous Systems
We are on the threshold of being able to place sensors everywhere. This has been precipitated in part by the continued rapid advances in sensor technology, which will allow us to embed sensors ranging from the nano-scale to the macro-scale on almost any physical device, at an economically viable cost.
Fortunately, computing and communications technology have been keeping pace with sensor technology, and all the ingredients are there for major breakthroughs in the near future in how we interface to, and control, our environment.
Serious challenges, however, must be overcome.
One of the most significant of these is the present difficulty in making appropriate decisions based on distributed information across a distributed network. To put this in context, it is well known that two simple dynamic systems can exhibit comparatively complex behavior when interconnected; the present challenge is to effectively design and control systems with many interconnected components.
Part of our research efforts are aimed at developing new tools for designing and controlling systems such as these. The emphasis is on tools for systems governed by differential and difference equations, both linear and nonlinear, with a large number of components, and interconnected through networks of structured connectivity. We seek to exploit the connectivity aspect of the problem as much as possible. Examples are varied and include regular interconnection structures for systems defined on lattices, and sparse structures for systems with limited connectivity such as vehicle platoons, ''smart'' materials with embedded actuation, aircraft flying in formations, and power distribution systems. Semi-definite programming algorithms can be brought to bear on these problems, resulting in computationally tractable algorithms for system analysis and control design. Other tools include Optimal Control to create motion primitives, Adaptive Control to improve system performance over time and to cope with changing conditions, Mixed Integer-Linear Programming to design cooperative strategies, and Distributed Estimation to build models of the environment from multiple, error-prone sources.
The underlying architecture of these systems is crucial to their success. To be effective, they must be modular, easy to adapt, and allow a large number of individuals to concurrently develop them. This is why, from a pedagogical perspective, we have adopted a multi-disciplinary team-based approach for many of our projects: individuals learn how to create modular subsystems that can easily interface with the subsystems created by other members of their team, and in the process acquire a solid understanding of feedback, dynamics and control.
This kind of ‘building block’ approach – where each self-contained subsystem can be easily put to use by non-experts – is crucial to effective systems engineering, where individuals across many fields must collaborate, where manufacturability and maintainability are key, and where prediction can greatly simplify the interface between the robots and the high-level algorithms that ultimately control them.
In today’s world, engineering, science, and mathematics are essentially utilitarian, and research in these areas is expected to have direct societal relevance. Unfortunately, «utilitarian» often means‚ «for the benefit of consumerism», and narrow metrics are typically used to gauge societal relevance.
We have an incredible opportunity to push the boundary of what is possible with control algorithms in the broadest sense when we remove the purpose-driven objectives typical to engineering from our research agenda. Novel ideas are often discovered in an unrestrained environment, and to encourage ‘out-of-the-box thinking’, we bring creativity to our research by building dynamic art installations for public display.
Our efforts are geared towards using motion design to explore the interface between mathematics, physics, engineering, and art. One of our research aims is to augment model-based control design with learning and adaptation to provide a flexible methodology for designing high-performance, robust systems. In the process, students are exposed to Systems Engineering, with an emphasis on system analysis, design, and integration. They learn skills such as requirements-driven design, manufacturability, maintainability, modeling and simulation of dynamic systems, and acquire an understanding of the interplay between system design, control design, and simulation.
- Prof. Raffaello D'Andrea