In this competition, one team is allowed to use one aerial vehicle, and use the vehicle to herd 10 roombas in the field. Each roomba is equipped with a dumper sensor on front and a lever sensor on top so that the vehicle can activate the sensors by landing in front of and also on the roomba. There are four obstacle roombas with a tall pylon, and we lose points if we hit these obstacles. The team that herds more roombas to the goal line wins. When a roomba crosses any boundary line of the field, it is removed from the field, so certain number of the roombas cross the goal line because they are randomly moving around; however, a roomba that is not touched by an aerial vehicle is not counted even when it crosses the goal line. This competition is held indoors, and the mission has to be done with a completely autonomous vehicle. For this very challenging competition, I first created a 6-DOF simulation to figure out the performance of a vehicle required for this mission. This simulation is used to test some strategies. The Youtube video above shows the simulation where an aerial vehicle lands in front of a roomba to change the direction of the roomba and navigate it toward the goal line. Another task of mine was to develop a vision system that can detect and track a roomba. I used the Viola-Jones detection framework to pull object proposals, and these proposals were filtered out by template matching and Kalman filter. Georgia Tech Aerial Robotics (GTAR) club participated in the competition and won the third best 2017.