Mobile Robot Navigation and Control: A Case Study
Nicholas Roy, Gregory Dudek, Michael Daum
Research Centre for Intelligent Machines
McGill University
Montreal, Quebec, Canada
Introduction
mobile robotics. The environment was produced as a result
of the recognition that progress in mobile robotics entails a
Robotic systems (and in particular mobile autonomous
progression from basic implementation of simple routines,
agents) embody a complex interaction of computational
through to the development of efficiently-implemented al-
processes, mechanical systems, sensors, and communica-
gorithms. With these considerations in mind, we have con-
tions hardware. System integration can present significant
structed a control and development interface for mobile
difficulties to the construction of a real system, because the
robotics experiments that permits a single robot to be con-
hardware is often built around convenience of design rather
trolled and/or simulated using any combination of manual
than convenience of system integration. Nonetheless, in or-
experimentation, simple automation and high-level algo-
der for robots to perform real-world tasks such as naviga-
rithms.
tion, localization and exploration, the different subsystems
of motion, sensing and computation must be merged into a
Implementation
single, realisable unit.
In the context of the AAAI competition, we are tapping this
Our group is investigating particular problems in the do-
infrastructure by rapidly constructing a set of client pro-
main of computational perception, in the context of mobile
cesses which embody task specific objectives for the meet-
robotics. In particular, we are concerned with environment
ing scheduling problem. The client processes group simple
exploration, position estimation, and map construction. We
sonar measurements into clusters used to classify regions of
have several mobile platforms integrating different sensing
the map according to a simple labelling hierarchy. By rec-
modalities, which we are able to control simultaneously
ognizing and following corridors in the environment, the
from a single source.
system travels between open spaces, or corridors using a
set of simple control heuristics.
Methodology
Our software tool allows us to transparently control and
To support this work, we have developed a layered soft-
simulate several different types of mobile robots. In addi-
ware architecture, that facilitates a modular approach to
tion, our work entails the use of a variety of sensing modal-
problems, in addition to building an abstraction of a robotic
ities, for example, sonar, laser-range, tactile sensing (Roy,
system (Dudek & Jenkin 1993). Our architecture involves
Dudek, & Freedman 1996) and video images. Furthermore,
three software layers: on-board real-time subsystems, off-
we have developed a customized video and range-sensing
board hardware-specific systems that abstract away hard-
platform called Quadris. The Quadris sensor can be used to
ware dependencies, and top-level “client” processes. This
further refine the labelling hypotheses generated from sonar
abstraction allows external software to interact with either
data.
a simulated robot and environment or a real robot complete
with sensors. The implementation is distributed across a
Long-term Development
network, and allows software to run on remote hardware,
Our long term objectives involve using these tools to ex-
thus taking advantage of specialized hardware available on
amine questions of spatial representation and exploration.
the network.
In particular, we have performed image-based position-
In appreciation of the necessity of simulation in addition
ing (Dudek & Zhang 1996), model-based localisation and
to real-robot control, we have developed a graphical envi-
exploration (MacKenzie & Dudek 1994), and topological
ronment for the development of algorithms and software for
map representation and exploration. We are also work-
∗
ing on extending this work to collaborative multi-robot ex-
The support of NSERC and the Federal Centres of Excellence
program is gratefully acknowledged.
ploration, with several agents performing independent ex-
