Evidential classification of dexterous grasps for the integration of perception and action
The objective of this article is to derive an algorithmic framework for the integration of perception and action in the intelligent control of multifingered robot hands. The two major goals include the integration of sensory information at multiple levels of specificity and the direct planning of grasp characteristics without searching through a task decomposition tree. To these ends, an evidential classification algorithm to determine the most appropriate grasp, given incompletely specified descriptions of an object and a manipulation task, is proposed. A supervised classification algorithm using fuzzy prototypes is developed. The algorithm is generalized for evidential classification, i.e. for problems where the input consists of belief functions rather than feature vectors in Euclidean space, for reasoning with uncertain data and knowledge in a simple robot manipulation environment. The algorithm is illustrated through a simple example. 85 references.
- Research Organization:
- George Mason Univ., Fairfax, VA (USA)
- OSTI ID:
- 6274456
- Journal Information:
- J. Robot. Syst.; (United States), Vol. 5
- Country of Publication:
- United States
- Language:
- English
Similar Records
Experiments in robotic sensori-motor control during grasp
Experiments in robotic sensori-motor control during grasp
Related Subjects
CONTROL SYSTEMS
ARTIFICIAL INTELLIGENCE
ROBOTS
ALGORITHMS
CLASSIFICATION
DESIGN
ERRORS
FINGERS
HANDS
OPTIMIZATION
PERFORMANCE
BODY
BODY AREAS
LIMBS
MATHEMATICAL LOGIC
990220* - Computers
Computerized Models
& Computer Programs- (1987-1989)