Duality for Convexification of Autonomous Control Problems
University of Texas at Austin, Department of Aerospace Engineering & Engineering Mechanics
Friday, October 17, 2014|
3:30pm - 4:30pm
About the Event
Many future engineering applications will require dramatic increases in our existing Autonomous Control capabilities. These include robotic sample return missions to planets, comets, and asteroids, formation ﬂying spacecraft applications, applications utilizing swarms of autonomous agents, unmanned aerial, ground, and underwater vehicles, and autonomous commercial robotic applications. A key control challenge for many autonomous systems is to achieve the performance goals safely with minimal resource use in the presence of mission constraints and uncertainties. In principle these problems can be formulated and solved as optimization problems. The challenge is solving them reliably onboard the autonomous system in real time.
Our research has provided new analytical results that enabled the formulation of many autonomous control problems in a convex optimization framework, i.e., convexiﬁcation of the control problem. The main mathematical theory used in achieving convexiﬁcation is the duality theory of optimization. Duality theory manifests itself as Pontryagin’s Maximum Principle in inﬁnite dimensional optimization problems and as KKT conditions in ﬁnite dimensional parameter optimization problems. Both theories were instrumental in our developments. Our analytical framework also allowed the computation of the precise bounds of performance for a control system, e.g., the bounds of agility for a vehicle, so that we can make accurate quantiﬁcation of the capabilities enabled. This proved to be an important step in rigorous V&V of the resulting control decision making algorithms.
This presentation introduces several real-world examples where this approach either produced dramatically improved performance over the heritage technology or enabled a new technology. A particularly important application is the fuel optimal control for planetary soft landing, whose complete solution has been an open problem since the Apollo Moon landings of 1960s. We developed a novel “lossless convexiﬁcation” method of solution, which will enable the next generation planetary missions, such as Mars robotic sample return and manned missions. Another interesting example is Markov chain synthesis with temporal and spatial safety constraints.
Behcet Acikmese is an Assistant Professor in the Department of Aerospace Engineering and Engineering Mechanics at the University of Texas at Austin. He received his Ph.D. in Aerospace Engineering from Purdue University. He joined NASA Jet Propulsion Lab•oratory (JPL) in 2003. He was a senior technologist at JPL and a lecturer in GALCIT at Caltech. At JPL, Dr. A¸cıkme¸se developed control algorithms for planetary landing, formation ﬂying spacecraft, and asteroid and comet sample return missions. He was the developer of the “ﬂyaway” control algorithms in Mars Science Laboratory, which successfully landed on Mars in August 2012, and the reaction control system (RCS) control algorithms for NASA SMAP mission, which will be ﬂown in 2014. Dr. Acikmese developed a novel real-time convex optimization based planetary landing guidance algorithm that was ﬂight tested by NASA JPL, which is a ﬁrst demonstration of a real-time optimization algorithm for onboard guidance of a rocket, showing dramatic performance improvements over the existing state-of-the-art techniques. Dr. Acikmese is an Associate Fellow of AIAA and a senior member of IEEE.
Contact: Ann Pace
Event Sponsor: Bosch, Eaton, Ford, MathWorks, Toyota, and Whirlpool
Open to: Public