IEEE Robotics & Automation Magazine - September 2011 - 105

There is also less control over illumination and background
scene clutter compared with stationary applications.
Movement Control
The eye-in-head configuration introduces uncertainties
from the changing views of moving eyes so that the
changes of illumination due to the environment and robot
motion are often commingled together, which cause difficulties for visual tracking and robot motion control [5].
Behavior Coordination
Current approaches need to maintain a centralized behavior network [2]. The difficulty appears to be that various
behaviors are merged together. There is no guarantee that
a goal can be reached. Modification of the network and the
prediction of consequences are both difficult.
Localization
The current approaches rely on the fusion of odometry with
onboard sensing based on a map or a bilateral technique
registering map data, i.e., simultaneous localization and
mapping (SLAM) [6]. Uncertainties involved in initial location, mapping, odometry, and sensing have a significant
influence on the accuracy and reliability of localization.
Usually, it is necessary to take a probabilistic approach to
identify the belief in an estimation, from a single hypothesis
to multihypothesis [1]. Poor localization precision may often
occur due to sensing errors or imperfect landmarks, e.g., for
global localization and kidnapped robot problems [7], [8].
Path Planning
Current approaches usually abstract a geometric environment into a topological map based on visual landmarks.
Planning is carried out on this topological map [1]. A
planned route has to be converted back to the geometric
space for continuous motion control. It introduces ambiguity and complexity in the linkages between the planning
and control spaces.
To overcome these difficulties, a system with cameras
distributed in the environment and connected by wireless
communication is presented in this article. It releases the
massive requirement for centralized computation into a
distributed camera network for the entire environment.
These cameras support the navigation of vehicles or
robots with a lower level of onboard intelligence. It is economically feasible in applications where a large number
of vehicles need to be controlled in a certain area, such as
navigation of wheelchairs in a care center and trading
environment intelligence for expensive onboard intelligence in every robot.
Wireless sensor networks provide the information
technology infrastructure for an intelligent environment,
which could lead to a new paradigm for navigation. If
intelligence can be pervasive in an environment [9], the
challenge of a complex environment, which causes the
reliability problem and hinders the broad application of

autonomous robots, will be greatly simplified. Initially, a
distributed sensor network can provide a topological map
of the environment. The routing to a geographic goal
causes querying a sensor sequence from the sensor network. The distributed sensors then become active landmarks for robot localization, which is more reliable and
efficient than onboard sensors. Finally, with a static
camera configuration mounted in the environment rather
than on a mobile robot, the intelligent environment will
facilitate both perception and control. Each sensor will be
in charge of a local region and therefore will face less
uncertainty and exhibit higher reliability. This current
research into wireless sensor network-based robot navigation took a hierarchical approach. Distributed sensors
were coordinated for high-level activities only, e.g., localization [11], routing [12], or event-driven behavior coordination [13]. A significant number of tasks of low-level
perception and control were left to the robots. This article
presents an efficient scheme for unifying routing, path
planning, trajectory generation, and motion control for
distributed wireless sensors. It is a complete navigation
solution developed in a project, "Wireless Mosaic Eyes to
Support Robot Navigation in an Intelligent Environment
(WiME)." Individual sensors in a wireless sensor network
usually have very limited fields of view, ad hoc communication, limited memory space, computational capacity, and
power onboard [9], [10]. This article includes two mechanisms for the coordination of distributed cameras with
limited onboard resources for robot navigation: multiple
Bloom filter-based routing and snake-based navigation.
System Architecture
With dispersed visual sensors, a prototype of the WiME system has been developed to support wheelchair and robot
navigation inside a building, as shown in Figure 1. The system consists of three hardware systems: 1) networked wireless visual sensors, 2) wireless-controlled robots, and 3) a
remote console.
The networked wireless visual sensors shown in Figure 1
are the main elements that implement image processing,
localization, and robot navigation. They are developed from
the Intel IMote2 main microcontroller board. The Intel
IMote2 is an advanced sensor network platform with builtin IEEE 802.15.4 [27] radio transceiver CC2420. The sensor
board is developed with an OV7620 visual sensor module,
which is a highly integrated interlaced/progressive scan
CMOS digital video chip with a resolution of 640 3 480. The
visual sensor board is connected with the IMote2 main controller through direct memory access (DMA) to enable highspeed data throughput.
In the project, we modified an off-the-shelf wheelchair
to be controlled by an Atmega128L processor. It receives
commands from the wireless visual sensors, measures
motor speeds from two encoders attached to the wheel
shafts, and drives and steers in response to commands, as
shown in Figure 1. It also connects with sonar bumper
SEPTEMBER 2011

IEEE ROBOTICS & AUTOMATION MAGAZINE

105

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - September 2011