This paper overviews the PLAYBOT project, a long-term, large-scale res
earch program whose goal is to provide a directable robot which may en
able physically disabled children to access and manipulate toys. This
domain is the first test domain, bur there is nothing inherent in the
design of PLAYBOT that prohibits its extension to other tasks. The res
earch is guided by several important goals: vision is the primary sens
or; vision is task directed; the robot must be able to visually search
its environment; object and event recognition are basic capabilities;
environments must be natural and dynamic; users and environments are
assumed to be unpredictable; task direction and reactivity must be smo
othly integrated; and safety is of high importance. The emphasis of th
e research has been on vision for the robot this is the most challengi
ng research aspect and the major bottleneck to the development of inte
lligent robots. Since the control framework is behavior-based, the vis
ual capabilities of PLAYBOT are described in terms of visual behaviors
. Many of the components of PLAYBOT are briefly described and several
examples of implemented sub-systems are shown. The paper concludes wit
h a description of the current overall system implementation, and a co
mplete example of PLAYBOT performing a simple task. (C) 1998 Elsevier
Science B.V.