11.1.   INTRODUCTION

Autonomy implies a degree of self-regulation inherent in a system’s operation,
and it is a key technology for future high-performance and remote
operation applications. The ‘‘promise’’ of autonomy includes (1) a reduced
need for human intervention, (2) increased performance range and capabilities,
(3) extended operation life, and (4) decreased costs. Examples of
important future applications that will rely on developments in the area of
autonomy and control include satellite clusters, deep space exploration, air
traffic control, and battlefield management with multiple uninhabited aerial
vehicles (UAVs) and ground operations. For these applications, an autonomous
architecture requires many online functions such as missionrtrajectory
planning, control reconfiguration, fault detection, and traditional lower-level
control loops. In addition, effective human command of many vehicles
requires information flow and fusion at several levels to facilitate (semi)auto-
nomous operation.

The concept of autonomous control is typically described as a layered,
feedback-based architecture with ‘‘degrees’’ of autonomy [1]. The approach
here is to focus on two lower levels of this architecture, as shown in Figure
11.1. This typical future autonomous vehicle architecture includes the following:
Low- and mid-level control including feedback compensation and trajectory
generation; fault detection to locate and recover from typical failures
within the system; and active state models that provide accurate model based
information to the other software-enabled control (SEC) components so that
they can perform adequately. SEC brings software and control algorithms
together to enable new functionality within the autonomous control paradigm.
Traditional feedback control techniques have few accommodations for maintaining
stability and high performance while operating in highly uncertain
environments and in the presence of failures or damage. SEC is designed to
deliver enabling functions such as real-time fault detection and control
customization, both of which will require models to accurately and reliably
fulfill their goals.

Most control and fault detection methods are model-based; accurate and
reliable modeling will allow these new autonomous control developments to
move past the low/no-noise and full-state feedback restrictions on to a
realistic implementation. For example, low-level reconfigurable control [2]
typically requires both a point model and an uncertainty set to guarantee
stability. Trajectory generation, such as those methods developed in the SEC
progam [3], use online model predictive approaches that require nonlinear
models with bounded uncertainties.