Robot helicopter learns autorotation emergency procedure for landing

Andrew Ng's autonomous helicopter group is at it again. After using reinforcement learning to teach their robotic helicopter how to fly inverted, they have now also managed to teach it the autorotation emergency procedure for safely landing after an engine failure. This procedure is used by real-life pilots to safely land their helicopter using the craft's potential energy to maintain sufficient rotor speed for landing.

Whereas during powered flight, the rotor drag is overcome by the engine power, during autorotation the rotor drag is overcome by the airflow through the blades. Effectively the potential energy of the helicopter (corresponding to its altitude) is transferred to rotor speed. This rotor speed allows the pilot to control the helicopter throughout its descent, and then slow down the helicopter before touching down.

Last month, the Stanford team presented a paper at the International Symposium on Experimental Robotics (ISER) describing their method. The paper also won the Ben Wegbreit IFRR Student Fellowship Award. Needles to say, autonomous control of a helicopter is far more difficult than the control of fixed-wing aircraft or ground vehicles. Especially in the case of landing with complete engine failure, the helicopter pilot has only one chance of successfully executing the autorotation maneuver and avoiding disaster. During the last few years, the Stanford team has made huge contributions towards solving the problem of helicopter autonomous control.

There are several videos of the autonomous helicopter in action at the team's official web site here.