RoboNexus Day 1 - Keynote: Matt Mason

Today's second keynote address, by Matt Mason, of the CMU Robotics Institute, was quite amazing. He showed off several of CMU's robotics related projects. The robotics department at CMU is primarily research focused and has 65 faculty.

Mason calls robotics, "where computing meets the real world" and takes a very broad view of things, breaking down the key 'robotic' components into "Perception," "Intelligence", and "Action." The two examples he showed us from the 'perception' category was a project that recognized and marked human facial featuers at a blazing 280 fps, and even maintained over 40 fps when the image was occluded. He also showed us video from a project that takes a 2D image and using machine learning algorithms, finds the 3D planes in the image and maps the image to those planes, allowing you to view the formerly 2D scene in 3D. He called it "Automatic Photo Pop-Ups." It was quite stunning.

Continuing in the same stunning vein, he showed us some video from a collisin detection and deformation program. Using a model of a plastic lawn chair, the computer processed 3,600 of them falling out of the sky and colliding - 1.2 billion collisions in total, all in real time.

Mason also had a video clip demonstrating path-finding algorithms. The group working on the project had digitized human movement (running, jumping, crawling) and then placed a 3D model of a human into an environment. The path planning algorithm then used these set of movements to navigate around the environment. It could also process multiple "people" moving around in the same environment, and even could do path finding with moving obstacles, like rolling boulders.

Mason concluded his talk with slides on the Dragon Runner, Gladiator, and the hybrid invertable ground vehicle. He says, "the future is in defining robotics broadly."

Thoughts?