Adaptive Fault Tolerant Control Allocation Strategies for Autonomous Overactuated Vehicles Alessandro Casavola ∗ Emanuele Garone ∗∗ ∗ casavola@deis.unical.it - DEIS, University of Calabria ∗∗ egarone@deis.unical.it - DEIS, University of Calabria Abstract: This paper presents a preliminary adaptive control allocation scheme for overactuated autonomous vehicles that is fault-tolerant with respect to actuator faults or loss of effectiveness. The main idea here is to use an ad-hoc online parameter estimator coupled with an allocation algorithm to perform on-line control reconfiguration. A simple algorithm is proposed for nonlinear discrete-time systems and its main properties are summarized for the disturbance-free case. Its effectiveness is also investigated by means of numerical simulations on an underwater vehicle. Keywords: Control Allocation, Adaptive Control Allocation, Control Reconfiguration, Autonomous Overactuated Vehicles, Underwater Vehicles 1. INTRODUCTION Actuators and sensors redundancy is an important issue to deal with for increasing the fault-tolerance of autonomous vehicles. It is a very common matter in airspace and under- water applications where, due to safety and performance reasons, a bank of redundant actuators and sensors is often used. Moreover, it is also common in gas, water and electrical distribution and production networks, where such a physical redundancy may be exploited to avoid interruption of services. In this paper we focus on the problem of control alloca- tion for overactuated systems (i.e. systems with physical actuator redundancy). One traditional way to solve it is to use multivariable optimal control design methods (Kwakernaak, 1972). Such an approach achieves both reg- ulation and control allocation at the same time. A different approach consists of using a simpler control law that spec- ifies the total control effort that has to be produced and in separately solving the so-called Control Allocation Prob- lem (CAP) i.e. the one of optimally distributing the desired total control effort over the available actuators. Due to its relevance, especially in flight control systems, CAP has been deeply investigated in the last decade and several methods have been proposed: Daisy Chaining (Buffington, Enn, 1996), Direct Control Allocation (Durham, 1993), Convex Optimization Based algorithms (Durahm, 1998; Bo˘ skovi´ c et al., 2002; H¨ arkeg˚ ard, 2002; Petersen, Fos- sen, 2005; Johansen et al., 2005) and PseudoInverse- Redistribution (PIR) methods (Bodson, 2002; Jin, 2005). A key necessity in the development of overactuated vehic- ular systems is to dispose of methods which exploit such a physical redundancy in the design of effective recon- 1 This work has been supported by MIUR Project Fault Detection and Diagnosis, Control Reconfiguration and Performance Monitor- ing in Industrial Process. figurable control strategies capable to avoid, or at least mitigate, the effects of actuator failures. A popular way to ensure some level of control reconfiguration is the use of adaptive control laws (Bodson, Groszkiewicz, 1997; Tao et al., 2002). An alternative approach that will be inves- tigated here is the so-called Reconfigurable Control Allo- cation (RCA) problem (Buffington et al., 1998; Bolender, Doman, 2005). The key idea is depicted in Fig. 1 where supposedly the control law has been designed on the basis of a virtual system with a minimal number of inputs v(t), fully equivalent to the physical inputs u(t) in generating a desired total control effort. Then, an allocation unit distributes at each time t the control v(t) on the physical actuators u(t), with a law that is allowed to be time- variant. Then, in the case of actuator faults, control re- configuration is possible in many cases without altering the control law by simply modifying the distribution of the total control effort v(t) to the remaining no-faulty actuators in u(t). This does not perturb in principle the closed-loop system dynamics because there are several way to distribute the control amongst actuators, all of them making the system behaving in the same way. In this paper a preliminary adaptive control allocation scheme is proposed able at solving RCA problems for non-linear discrete-time systems. Such systems are subject to actuator faults or loss of effectiveness. Unlike other works on the topic, here the algorithm is not assumed to know the occurrence of a fault. On the contrary, an adaptive mechanism is used to estimate possible loss of effectiveness and make possible the on-line computation of the allocation rules by solving a standard constrained QP problem. The main properties of the scheme are summarized. For simplicity all developments are done in a disturbance-free scenario without considering model uncertainty. All issues related to the robustness properties of the algorithm are demanded to future studies.