SUBJECT: Ph.D. Proposal Presentation
BY: Hannes Daepp
TIME: Friday, February 28, 2014, 2:00 p.m.
PLACE: MARC Building, 114
TITLE: Model Predictive Control for High Performance Pneumatic Actuation
COMMITTEE: Dr. Wayne Book, Chair (ME)
Dr. Aldo Ferri (ME)
Dr. Nader Sadegh (ME)
Dr. Mark Costello (AE)
Dr. Eric Barth (Vanderbilt, ME)


Pneumatic actuators possess a number of qualities that make them potentially versatile actuators: they have high power and force density, are clean, safe, and low cost, and possess inherent compliance and potentially adaptable stiffness that make them useful for contact and interaction tasks. However, control of pneumatic actuators has proven difficult, limited by the inherent compliance of the actuator, friction in the cylinder, and third order dynamics that are both nonlinear and discontinuous. In general, past controls solutions have had limited application, applied where position tracking is not critical, or using high-gain PD controllers to transform the system into a stiffer one that succeeds in precision tracking but possesses high output impedance and lacks compliance. Of the variety of advanced controllers that have been tested, the most successful have been sliding mode controllers, which add robustness criteria to simpler feedback controllers. Input shaping has shown that the tracking problem can be solved in the open-loop case without sacrificing compliance and impedance goals. Model predictive control offers the potential to extend this concept to the pseudo-closed-loop case, by effectively iteratively solving a feedforward problem to achieve good tracking performance of a system with high compliance and low output impedance that interacts safely and securely with its surroundings. A predictive observer can then be used to compensate for friction and especially stiction proactively, rather than using an additive compensation term at each instant. This may improve performance for a pneumatic system, which has generally slow dynamics. Finally, the predictive control enables the user to place constraints on the optimization, thereby enabling the controller solution to operate closer to the optimal capability of the system, which is subject to dynamic and mechanical limits. The resultant system resembles the increasingly common series elastic and variable stiffness actuators, coupled with a new method to achieve control of compliant actuators for use in robots that require both good tracking and environmental interaction.