Objective 1: Unsupervised Nonlinear System Identification
Within the ACCOST project, motivated by previous work on prescribed performance control, we aim at designing: (i) an on-line RBF neural network identification scheme; (ii) a PE reference trajectory and (iii) a controller generating the appropriate control input signals that guarantee the satisfaction of PE condition for the RBF regressors employed. As a consequence, the neural network weight estimates would be not only uniformly bounded but also would converge to small neighborhoods of their actual values; thus achieving the goal of learning the actual system nonlinearities with quality guarantees.
Objective 2: Control under Actuation Limitations
Within the ACCOST project and motivated by previous works on prescribed performance control with deadzone, backlash and hysteresis input nonlinearities, we aim at developing a novel adaptive modification of the Prescribed Performance Control methodology that adjusts online the performance specifications when the input constraints tend to be violated, i.e., when a contradiction between what is desired and what is feasible arises. In particular, the ultimate goal will be to find out how input constraints affect the output performance through the system dynamics, which still remains open in the related literature since no constructive method has ever been proposed that achieves the best feasible performance under certain actuation constraints.
Objective 3: Control under Feedback Limitations
Within the ACCOST project and motivated by previous works on prescribed performance control, we aim at designing a novel observer scheme, reinforced with predefined performance attributes without either resorting to the high-gain approach or incorporating a priori knowledge of the actual system nonlinearities. Therefore, the intriguing properties of Prescribed Performance Control methodology would be sufficient to establish output feedback control with prescribed performance via simply securing the boundedness of the closed-loop system trajectories, thus resulting in significantly less complex and more robust design, compared to existing works in the related literature.
Objective 4: Demonstration in Aerial Robotics
It is very important to demonstrate and validate the overall control framework that will be designed in the context of the ACCOST project. For this reason, we aim at multiple experimental and realistic simulation studies that will incorporate all aforementioned theoretical aspects for a variety of application domains in aerial robotics. In particular within ACCOST, we shall consider multi-copter aerial robots involved in three case-studies: (i) trajectory tracking in free space, (ii) contact force/position control (in such case the multi-copter will be enhanced with a manipulator for increased intervention capabilities) and (iii) multi-robot cooperative formation. The studies should involve: (a)realistic modeling of the complex dynamics (including external disturbances and measurement noise) in each case-study, (b)an autonomous system identification module to extract an accurate dynamic model, (c)a control scheme that takes into account actuation as well as feedback constraints and guarantees certain predefined performance requirements imposed by the operational specifications of each case separately, (d)an automatic tuning process for the control parameters and (e)extensive comparative results with the current state of the art algorithms that are employed in each case-study, in order to put forward the full prospect of the proposed control framework.
