Multi-sensor input path planning for an autonomous ground vehicle

Abstract

Autonomous Unmanned Ground Vehicles (UGV's) are mobile platforms that serve a wide range of specialized applications in urban, military, domestic, and industrial settings. UGV's can be remotely operated or autonomous and usually include a variety of sensors and manipulators that are used to solve specific investigation tasks. They also include sensory input for use in autonomous navigation algorithms. For example, radio activity or explosive sensors can help the remote assessment of a dangerous area. In the case of autonomous navigation, a UGV usually employs Light Detection and Ranging (LIDAR) sensors a. k. a. laser range finders, ultra sonic sensors, cameras, and Global Positioning System (GPS) receivers to avoid obstacles and follow a set of GPS waypoints that define a path for the UGV to cover. This paper presents the development of autonomous path planning in a UGV that uses various sensors including, a laser range finder, a digital compass, a GPS receiver, and computer vision. The sensor data is fused in a “cost matrix” that assigns positive numerical values to obstacles detected using the various sensors. Negative value contributions are added to the cost matrix is areas corresponding to the desired heading dictated by GPS waypoint navigation. A cost function is implemented by adding cost matrix values over several possible paths crossing the matrix, causing the lowest-cost path to be selected as the UGV's next heading. The algorithm was tested on a grassy path with white lines defining an allowable path that includes various physical obstacles.