A perception system for detecting brake levers in outdoor rail yard environments

摘要

A rail yard is a dangerous environment for humans to work in, primarily because of the possibility of serious injuries associated with moving rail cars, locomotives, and uneven terrain. For robots to act autonomously in such environments, there exists a need for a perception system that can act reliably under uncertain conditions. This uncertainty arises from multiple factors—uncontrolled lighting, uneven terrain, variability in appearance of objects, dynamic environment, noisy sensors, and controllers. In this paper, we present a perception system for a mobile robot that leverages information from various sensing modalities to act reliably in a partially observable environment. We show how our system can detect brake levers on a rail car, by fusing information from multiple detectors. We validate our approach by performing tests in an actual rail yard over multiple days and nights.