Geometric Sensor Fusion in Robotics

Abstract

To operate effectively in increasingly complex environments, robots must be capable of integrating information from many different sensory sources, into a robust and consistent description of an operating domain. We describe one method of achieving this goal which starts from the premise that every sensor and algorithm can be treated as a “geometry extractor.” Both the observations and the sensors themselves are described in terms of uncertain geometry. This geometry provides a common language by which disparate sensors can communicate information, different observations compared, verified and integrated.