Occasionally we see robots that are so awesome that we wonder why they haven't conquered the world yet. So it is with the newest unmanned aerial vehicle (UAV) from UPenn's GRASP Lab, which autonomously performs "precise agressive maneuvers" in a great video (see below). 'Agressive' is the key word here as the quadrotor drone sounds like a swarm of angry hornets as it bursts through a series of windows at different angles and orientations. It also performs flips, weaves through other UAVs, and perches on walls with Velcro. This thing is amazing!
Quadrotors provide a good deal of stability and maneuverability and make ideal spy drones. MIT is developing an iPhone control system for quadrotors for military use and Parrot released a retail version (also controllable via iPhone) with augmented reality applications. GRASP's version of the device, however, can go where no drone has gone before - ducking through openings at high speeds with just three inches of clearance. That's an incredible degree of autonomous maneuvering and it shows how well these UAVs may be able to adapt to urban environments and unconventional terrain.
The UPenn quadrotor is the work of grad student Daniel Mellinger, his advisor Vijay Kumar, and research scientist Nathan Michael. They used 20 vicon cameras to digitally map the quadrotor in its space. You can clearly see the white sphere markers on the UAV and the 'windows' in the video. Combined with data from the quadrotor's IMU, the visual tracking allowed the UAV to perform its precise and aggressive maneuvers. While that tracking system provided amazing results, it's obviously not something you want to have to set up in the field. It will take considerably more research to allow the UAV to perform the same maneuvers with its own instrumentation in a non-mapped space. In other words, as aggressive as the GRASP quadrotor may be, you don't have to worry about it flying through your window anytime soon. That's definitely a load off my mind. Now if I could only find a way to avoid an unstoppable robot dog...
[image credit: Daniel Mellinger/GRASP UPenn]
[source: GRASP UPenn]