Flying iPhone toy looks ripe for AI research
AR.Drone is a flying iPhone accessory, with enough on-board sensors and processing to wet the appetite of even the most jaded artificial intellegence researcher.
Named from ‘augmented reality drone’, the four-bladed electric helicopter comes from French firm Parrot, better known for Bluetooth headsets and chic wireless loudspeakers.
What separates it from traditional radio-controlled aircraft is a layer of on-board intellegence that takes simply forward/rotate/up commands from the operator and turns them into signals for the four on-board motors.
And this is closed-loop control, with on-board sensors used for feedback – to the extent that the user command set includes: hover, land, and take-off.
The embedded computer is based around a 468MHz ARM9 running Linux with 128Mbyte of 200MHz DDR.
The aircraft includes its own b/g Wi-Fi master and works over the radio link with iPhone, IPod Touch or iPad.
Communication is two way, with images from AR.Drone’s two cameras – one forwards and one down – relayed back to the controlling IPhone.
The image from the 60frame/s CMOS downward camera is also processed to provide lateral feedback for the hover command, with an Rx-Tx pair of 40kHz ultrasonic transducers included to give height feedback up to 6m altitude.
AR.Drone gets its augmented reality title from games played between operators of several drones in the same environment.
In this case, the image from the 640×480 (VGA) forward camera is processed to automatically identify carry different coloured tags on the other drones in the game and can estimate distance if it is between 0.3-5m.
Using this data the phone overlays shoot-em-up graphics over the video image. Working with software on the controlling phone.
The inertial guidance package is a three axis accelerometer, a two axis gyro, and a one axis yaw gyro.
The accelerometer was specified with the help of component distributor Acal.
For the inertial guidance system, a Bosch BMA150 3-axis accelerometer was used in conjunction with a 2-axis gyro and a 1-axis yaw gyro to provide the data for the auto-pilot. Initially the AR.Drone used a QFN-packaged accelerometer, but then an LGA-packaged version was used to reduce both the power consumption and cost.
The four 35,000rpm brushless dc motors are powered from a three cell 11.1V 1Ah lithium polymer battery which lasts for around 12 minutes of flight.
Forward speed is 5m/s (11.2mph).
The drone weighs 380g in its more aerodynamic ‘outdoor’ form, which increases to 420g when its indoor hull is added -a hull which puts bump protectors around each of the lift propellers.
The indoor hull increases dimensions from 45×29 to 52.5×51.5cm.
Since the demise of Sony’s AIBO robot dog, the artificial intellegence community has been deprived of a nimble low-cost (relatively) platform with plenty of on-board sensing and processing capability.
AR.Drone, for which a software development kit is available, could become the new AIBO.
For fans of four rotor flight who want even more customisation potential, Aeroquad is a DIY project based around the Arduino embedded computer which has gyro-stabilised flight through built-in control loops for roll, pitch and yaw.