Scott A. Smolka
Stony Brook University, USA
V-Formation as Optimal Control
(joint program CONCUR & FORMATS)
Abstract and Bio »
Abstract: In this talk, I will present a new formulation of the V-formation problem for migrating birds in terms of model predictive control (MPC). In this approach, to drive a flock towards a desired formation, an optimal velocity adjustment (acceleration) is performed at each time-step on each bird's current velocity using a model-based prediction window of T time-steps. I will present both centralized and distributed versions of this approach. The optimization criteria used is based on fitness metrics of candidate accelerations that V-formations are known to exhibit. These include velocity matching, clear view, and upwash benefit. This MPC-based approach is validated by showing that for a significant majority of simulation runs, the flock succeeds in forming the desired formation. These results help to better understand the emergent behavior of formation flight, and provide a control strategy for flocks of autonomous aerial vehicles. This talk represents joint work with Radu Grosu, Ashish Tiwari, and Junxing Yang.
Biography: Scott Smolka is a professor in the Department of Computer Science at the State University of New York, Stony Brook. He is also president and co-founder of Reactive Systems, Inc. He has been on the faculty at SUNY at Stony Brook since 1982. Smolka holds an A.B. and A.M. in Mathematics from Boston University and a Ph.D. in Computer Science from Brown University. His work focuses on analysis techniques for reactive systems and he also has extensive experience in building verification tools, including Winston and the Concurrency Factory.
Ufuk Topcu
University of Texas at Austin, USA
Adaptable Yet Provably Correct Autonomous Systems
(joint program FORMATS)
Abstract and Bio »
Abstract: Acceptance of autonomous systems at scales at which they can
make societal and economical impact hinges on factors including how
capable they are in delivering complicated missions in uncertain and
dynamic environments and how much we can trust that they will operate
safely and correctly. In this talk, we present a series of algorithms
recently developed to address this need. In particular, these
algorithms are for the synthesis of control protocols that enable
agents to learn from interactions with their environment and/or humans
while verifiably satisfying given formal safety and other high-level
mission specifications in nondeterministic and stochastic
environments.
We take two complementing approaches. The first approach merges data
efficiency notions from learning (e.g., so-called probably approximate
correctness) with probabilistic temporal logic specifications. The
second one leverages permissiveness in temporal-logic-constrained
strategy synthesis with reinforcement learning.
Biography: Ufuk Topcu joined the Department of Aerospace Engineering and Engineering Mechanics at The University of Texas at Austin as an assistant professor in Fall 2015. He received his Ph.D. degree from the University of California at Berkeley in 2008. Before joining The University of Texas, he was with the Department of Electrical and Systems Engineering at the University of Pennsylvania. He was a postdoctoral scholar at California Institute of Technology until 2012.
Abstract and Bio »
Abstract: This talk provides a retrospective look at the past, present,
and future of speed scaling systems. Such systems have the
ability to auto-scale their service capacity based on demand,
which introduces many interesting tradeoffs between response
time (a classic performance metric) and energy efficiency
(a relatively recent performance metric of growing interest).
The talk highlights key results and observations from the past
two decades of speed scaling research, which appears in both the
theory and systems research communities. One theme in the talk
is the dichotomy between the assumptions, approaches, and results
in these two research communities. Another theme is that modern
processors support surprisingly sophisticated speed scaling
functionality, which is not yet well-harnessed by current
algorithms or operating systems.
During the stroll, I will also share some insights and
observations from our own work on speed scaling designs,
including coupled, decoupled, and turbo-charged systems.
This work includes analytical and simulation modeling, as
well as empirical system measurements. The talk closes with
thoughts about future opportunities in speed scaling research.
Biography: Carey Williamson is a Professor in the Department of Computer Science at the
University of Calgary. His educational background includes a B.Sc.(Honours)
degree in Computer Science from the University of Saskatchewan in 1985,
and a Ph.D. in Computer Science from Stanford University in 1992.
Dr. Williamson's research interests include Internet protocols, wireless
networks, network traffic measurement, workload characterization, network
simulation, and Web server performance. From 2001-2011, he held an iCORE
(Informatics Circle of Research Excellence) Chair in "Broadband Wireless
Networks, Protocols, Applications, and Performance", as well as an NSERC
Industrial Research Chair in "Wireless Internet Traffic Modeling" from
2004-2009.
He is a member of ACM, IEEE Computer Society, and IFIP WG 7.3. He served
as SIG Chair for ACM SIGMETRICS from 2007-2011, and is currently the
co-Editor-in-Chief of the ACM Transactions on Modeling and Performance
Evaluation of Computing Systems (ToMPECS).