|The CareBot presently uses
an 85Kbps RF modem data link to its multiple MCU's on board and a
CPU brains off board system architecture with usage of low priced
IEEE802.11b (wireless networking). This approach enables obsolete
CPUs to be physically and cost effectively removed from the personal
robot platform with numerous beneficial implications to the cost of
a personal robot. The minimal hardware platform requirements of locomotion,
power, and sensor systems are sufficient for a high level of autonomy
in many homes and work places and are achievable at consumer electronics
prices of a few thousand dollars.
With many of the traditional hardware cost issues
significantly reduced, we now turn to the ninety percent problem in
robotics: the software. Even with the seemingly unlimited computational
power now available to the mobile robot with low cost RF data links
and PC's, the difficulty of writing software that enables high levels
of autonomy should not be underestimated.
Multiple sensor systems are needed to enable high levels
of autonomy. We perceive our own world using multiple sensor systems
that coordinate and complement our overall perception. Just as a blind
person, without sensory aids such as a cane, (whether physical or
ultrasonic,) does not have the level of autonomy as most sighted persons,
this is true for mobile robot sensor systems. Humans use sight, sound,
touch, smell, and even infrared routinely in our daily movement around
our homes and workplaces. When we step on something, such as a soda
bottle, our tactile sense warns even the sighted to take a closer
look. Traditionally, nested if/then/else statements are used to program
dozens of sensors to interact. When using more than fifteen or twenty
discrete range findings, and/or tactile sensors, the creation, maintenance
and extension of the software to intelligently interpret the increasing
number of managed data points becomes difficult. Generally, at some
small, finite number, this software and hardware interface becomes
unmanageable. A robust approach to sensor fusion would minimize, if
not eliminate this problem. Of the personal robot companies mentioned
in this discussion, only GeckoSystems has a sensor fusion offering
while most of the other robots have very low sensor counts.
Mapping the environment for path planning is another
impediment to the systems integration necessary for a cost effective
and utilitarian personal robot. Often referred to as the self-localization
("where am I?"), this allows real-time path planning (a
difficult software navigation issue). The map must be as accurate
and repeatable as possible. With low sensor data counts due to the
limits of traditional if/then/else programming, maps generated using
data available by most commercially available personal robots are
sparse and consequently lack sufficient detail for near collision
free running or patrolling.
Perhaps the greatest obstacle in utilitarian personal robot
development has been in near real-time path planning since traditional
approaches to path planning are incredibly computationally intensive.
Typical solutions for the last ten to twenty years use workstation
level computers, off line (not in real time), to determine a path
for the mobile robot to follow from the kitchen to the master bedroom.
Dr. Rodney Brooke's work at MIT, well known for its
"subsumptive robot behaviors", enables a mobile robot, for
example, to encounter and avoid an unforeseen obstacle while on path,
then re-orient, and resume the original path. This is sometimes referred
to as "cognitive navigation." The extremely
|| high computational requirement
for path planning still precludes the level of autonomy necessary
for a truly utilitarian personal robot. However, working independently,
CMU and GeckoSystems have developed similar, although different, "flood
fill" solutions to this decades old problem of near real time
path planning. The consumer will not wait twenty minutes while their
PCR determines how to move from the kitchen to the master bedroom.
The GeckoSystems solution only uses fifteen to twenty percent of a
commodity, 300Mhz, PC to achieve path planning in near real time.
The following analogy illustrates the potential impact of
the breakthroughs by CMU and Gecko-Systems. Until a few years ago,
it was believe audio compression had encountered a theoretical maximum.
Therefore, the average listener could not further compress audio files
since after decompression; the resulting music would not have sufficient
fidelity to ensure enjoyment. MP3 audio file compression and decompression
created a dramatic paradigm shift in music distribution within a few
The efficiency of the GeckoSystems' GeckoBrain fuzzy
hybrid software and PCR hardware architecture enables market penetration
as a robust personal robot platform with a high level of autonomy,
long battery life and expansion capabilities. Expansion can include
vacuuming, intruder/fire detection, plant watering, remote surveillance,
virtual visiting and a number of other areas. In regards to family
care, both CMU and GeckoSystems believe there is a great need for
a PCR in eldercare. Additional opportunities for AI augmentation for
RF linked PCR's are discussed later in this article.
Potential Tasks for the PCR
Let us look at the types of utilitarian yet cost effective
tasks now possible based on the software and hardware architectures
we just discussed.