Event Location: NSH 1507
Bio: Stephen Nuske is a project scientist in the Field Robotics Center at Carnegie Mellon University. He originally joined CMU as a PostDoc in 2008 after completing his PhD at the University of Queensland and the Commonwealth Science and Industry Research Organisation (CSIRO), Australia.

Abstract: Maps can store and convey many valuable attributes, though in some domains the challenges posed by the environmental conditions can preclude the possibility to generate accurate maps. Visually guided robots offer avenues to generate new and desirable types of maps that are not possible to create otherwise.
Rivers with significant canopy cover cannot be mapped from satellite; the sensory data must be collected from beneath the tree line. This talk presents a low flying Micro Aerial Vehicle (MAV) mounted with cameras combined with self-supervised scene segmentation and de-biased visual odometry algorithms designed to plot the width and course of riverine environments.
In vineyards and orchards, maps of expected harvest yield are highly sought after by growers. Conventionally, measurements would be manually taken with sparse sampling that leads to inaccurate and incomplete yield forecasts. A ground robot with cameras and texture-classification algorithms can segment fruit among the tree’s leaves and branches to capture the state of a grower’s yield with high-resolution and accuracy.
Results from the talk highlight that robots with visual sensing and novel algorithms can add value to industries in environments that traditionally were too difficult to map.