An Object-Based Representation for Multisensory Robotic Perception

J.C. Rodger and R.A. Browse

Abstract

The inclusion of multiple sensors in robot systems emphasizes the need for general techniques to combine data across sensor systems. This paper describes a simple, uniform, object-based representation and consistency scheme that facilitates the combining of input from an arbitrary number of sensors in order to recognize and locate objects. The approach assumes that raw sensor data yields features that invoke objects, together with associated orientation and location constraints. This makes it appropriate for model-based recognition systems using spatial sensors (eg. Photometric, tactile, ranging). The uniform representation permits application of simple consistency operations to extract interpretations that are supported by all sensor inputs. This approach has been tested in a demonstration system that simulates the combination of photometric and force array sensing in order to recognize and locate modeled objects.