Slashdot Mirror
Smart Flying Robots
Chernyakov writes "MARVIN, a fully autonomous RC helicopter built by the Technische Universitaet Berlin, won the
2000 International Aerial Robotics Competition.
MARVIN has a radio-linked ground station consisting of several networked Linux machines (which provide the computing power for vision, mapping and flight-course generation).
The robots' mission is to fly into a disaster area complete with fire, water and smoke hazards, to locate and avoid threats to itself, to find bodies, distinguish from survivors and the dead, identify hazardous materials containers, determine if the container contents are radioactive, biohazardous, or explosive (by reading the labels), generate a detailed map of the disaster area, photograph the area, and return safely back to base.
MARVIN pulled it off completely autonomously, with no human help or intervention. High quality (90 MB) and Low quality (12 MB) MPEGs of the robot are available."
"A brain the size of a Beowulf cluster, and what do they have me do? Fly around disasters taking pictures..." *sigh*
In order to fly autonomously (by itself), MARVIN uses a LOT of hardware: 3 accelerometers, 3 gyroscopes, 3 compasses (redundancy), 2 ultrasonic range finders, a differential GPS system (accurate to 2 centimeters), a fire detector, and an altimeter. It also has an onboard CPU, communications equipment, power distribution, etc.
.60-size XCell.
The helicopter itself looks like a standard
Anyone who has ever flown an RC helicopter knows how difficult it is for a human, much less a computer. The software usually involves use of a Kalmon Filter to fuse the sensors, and neural networks to build fuzzy control rules for the flight surfaces (aileron, rudder, collective, etc) in real-time.
The video, boring for some, shows the helicopter taking off and flying over the disaster area. The heli adjusts its location constantly for better views of targets then flies away and lands. There are good shots of the ground station showing Mission Control, robot-vision, and flight-path mapping. Also some good shots of not-so-successful flights and what can go wrong.
This is the tenth year of the IAR Competition; each year the mission gets more complicated. In 1991, autonomous flying vehicles (no hovercraft) simply needed to pick up a disc at one end of the field and drop it off at a deposit-point on the other end. No one completed the mission that year.
As a side point, let me just say that technology can always be used for evil, but development of robots such as these are most useful in what the industry terms D^3 (D-cubed) environments: Dull, Dirty, and Dangerous.
Dull: Flying over thousands of acres of forest painstakingly examining almost each tree for insect damage.
Dirty: A disaster area with potential exposure to lethal gases and the like.
Dangerous: Photographing a volcano for threat analysis.
I have a copy of the 90 MB mpeg -- I'll try to mirror it on Mojo Nation, but my one-way cable modem might not like it.