These talks will be self-contained presentations of mathematical concepts and theorems. They are intended to be accessible to all mathematics students.
 
 

Humanoid robots are a natural vehicle for human-machine cooperation in an office/domestic environment. Our research aims to explore the necessary building blocks for a humanoid robot to perform interactive manipulation tasks (grasping and placing objects) in this type of unstructured setting, with a focus on visual perception. An important component is the robot's ability to locate and classify various objects (cups, bottles, books etc) required to complete a given task. For this purpose, we developed a novel range data segmentation algorithm capable of identifying geometric primitives (planes, cylinders, cones and spheres) in the robot's workspace. Our split-and-merge segmentation algorithm is based on estimation of the number of non-zero principal curvatures and local convexity at each range data element. Experimental comparison with existing methods shows that our approach has greater robustness to noise and successfully models a variety of domestic objects. We will also present an overview of our related robotics research, including visually guided control of robotic limbs, robust range data acquisition and object tracking.