Page 431 - Automotive Engineering Powertrain Chassis System and Vehicle Body

CHAP TER 1 4. 2 Decisional architecture

14.2.2 Robot control architectures view of human reasoning, and it is often referred as the
‘sense-model-plan-act’ (SMPA) scheme. In practice, this
and motion autonomy
concept is implemented into robotic control systems
using a hierarchical architecture made of three main
14.2.2.1 Definitions and taxonomy components: perception (which includes sensing and
modeling functions), decision and action.
The development of robot control architectures consti-
tutes for engineers and scientists one of the most chal- Perception. Considered as a key feature in a robotic
lenging frameworks for integrating and testing intelligent system, the perception function may be seen as the
systems, inspired from attributes of living beings such as first basic function of a deliberative architecture. The
perception, interaction and reasoning. Robot control ar- main purpose of this first stage of the control process
chitectures are rather understood in terms of software is to construct a model of the environment from
architectures, and consequently are closer to domains sensory data and a priori knowledge (e.g. topological
related to computer science and control engineering. A or a grid-based models). This model is subsequently
basic definition for a robot control architecture can be used for planning robot actions and for checking that
found in Arkin (1998): ‘Robotic architecture is the the robot actions have correctly been executed.
discipline devoted to the design of highly specific and However, world reconstruction from sensory data is
individual robots from a collection of common software a complete active research domain, having already
building blocks.’ motivated a great number of research works and
The state of the art in this domain includes a large approaches; this research domain is still open.
number of approaches, sometimes guided by research work Decision. The second processing phase of a delibera-
on ethology and cognitive sciences. One of the most chal- tive architecture is referred as the decision module.
lenging domains for testing and evaluating these approaches, It consists in ‘reasoning’ about the task model and the
particularly when real-time constraints have to be verified, is environment model, in order to decide what is the
mobile robotics. This is why most of the significant contri- more appropriate sequence of actions to execute. In
butions in this research field come from work on mobile practice, this reasoning phase is often implemented
robot and autonomous guided vehicles. The next sections as an off-line motion planning task. This is why
outline the state of the art in mobile robot architectures, motion planning has been a very active research
using a commonly agreed taxonomy. This taxonomy is based domain for about 20 years.
on three main paradigms on which a large number of control Action. The last processing phase of a deliberative
architectures have been developed: architecture is to control the robot actuators in
order to execute the planned actions. Recent
The deliberative paradigm. In this approach, the research work in this domain addresses robust con-
system uses a model of the world – an a priori trol techniques and sensor-based control approaches.
known model, or a model reconstructed from sen-
sory data – in order to plan the actions that the robots The first robot control architectures reported in the lit-
have to execute. This approach leads to a sequential erature are based on such an approach. In particular, this
decomposition of the whole process, and to highly type of architecture has been used for controlling the first
hierarchical systems. mobile robot having a partial autonomy: the robot Shakey
The reactive paradigm. This approach is based on (Nilsson, 1984). This robot, designed at the beginning of
a tight coupling between sensors and actuators, for the 1970s at the Stanford Research Institute, used
continuously producing the required controls. This a video-camera as a sensor and was theoretically able to
approach usually relies on a decomposition of the move in a highly constrained environment. Its reasoning
system into elementary behaviours which can be capabilities were derived from problem-solving tech-
combined and executed concurrently. niques developed in the field of artificial intelligence. The
typical tasks that could be achieved by Shakey consisted
The hybrid paradigm. This approach consists in com- in finding a known object (i.e. an object described by its
bining the deliberative and reactive paradigms, in
shape and its colour) in a room, and in pushing this object
order to try to exploit the advantages of the two
previous approaches. Most of the current approaches up to a given point. Unfortunately, each simple move-
are of this type. ment of Shakey required more than one hour of external
computing, and it had a strong probability of failure at
execution time.
14.2.2.2 Deliberative architectures The architecture of the Stanford Cart developed at
the University of Carnegie-Mellon (Moravec, 1983)is
This approach relies on traditional paradigms of artifi- also representative of the deliberative paradigm. In this
cial intelligence. It tries to implement a simplified approach a 3D-vision system provided the robot with the

438