The term "marker" as used in the active vision and reactive planning literature is poorly defined. The underspecification of the marker concept limits the usefulness of this important tool for building perception and action systems for autonomous robots. This paper refines the concept of markers, both at the level of their role in perception-action system, and the level of the representations used in their implementation. Our definition encompasses markers in dynamic 3D environments, which are retained even when not in the current perceptual input, either due to occlusion, a limited field of view, or sensing errors. We develop methods for maintaining the representation in the face of these difficulties. The marker concept is further elaborated by a categorization of marker types we have found useful in our research. Relationships among markers are also discussed: relationships which potentially form a link between classical and reactive planning. Finally, we describe a marker-based implementation of an autonomous agent performing search and avoidance tasks in a three-dimensional virtual environment.
[1]
S. Ullman.
Visual routines
,
1984,
Cognition.
[2]
Ian Horswill,et al.
Polly: A Vision-Based Artificial Agent
,
1993,
AAAI.
[3]
Maja J. Mataric,et al.
Integration of representation into goal-driven behavior-based robots
,
1992,
IEEE Trans. Robotics Autom..
[4]
Dana H. Ballard,et al.
Reference Frames for Animate Vision
,
1989,
IJCAI.
[5]
Rich Gossweiler,et al.
Alice: a rapid prototyping system for building virtual environments
,
1994,
CHI Conference Companion.
[6]
Andrew McCallum,et al.
Overcoming Incomplete Perception with Utile Distinction Memory
,
1993,
ICML.
[7]
David Chapman,et al.
Vision, instruction, and action
,
1990
.
[8]
Rodney A. Brooks,et al.
Elephants don't play chess
,
1990,
Robotics Auton. Syst..
[9]
David Chapman,et al.
What are plans for?
,
1990,
Robotics Auton. Syst..