Aerei robot come antenne, un esperimento in Svizzera

Il Laboratory for Intelligent Systems della Scuola Politecnica Federale di Losanna sta studiando l'uso di "sciami" di minuscoli (sotto il mezzo chilo di peso) velivoli robotizzati (UAV) alimentati da un motorino a batteria di litio da dispiegare nelle aree dei disastri naturali e altre emergenze per realizzare infrastrutture di comunicazione volanti. Una rete cellulare in cielo a bassa quota. Ogni velivolo robotizzato ha una autonomia di circa mezz'ora e l'aspetto interessante riguarda gli algoritmi per il controllo di questi sciami, che richiedono un singolo punto di controllo ed evitano le collisioni dialogando continuamente tra loro tramite connessioni ZigBee. Due gli algoritmi sperimentati finora, uno basato sull'osservazione degli sciami di formiche che mediano e coordinano i loro movimenti rilasciando feromoni, l'altro sull'autoapprendimento delle macchine a controllo numerico.

Questo breve articolo da Embedded Internet Design fornisce le informazioni essenziali ma il sito del Laboratorio elvetico è molto più completo.

Swarm intelligence controls robotic planes

R. Colin Johnson

PORTLAND, Ore.–A new strategy for coordinated flight of unmanned aerial vehicles (UAVs) devised by the Swiss Federal Institute of Technology uses coordinated communications to allow a single operator to control an entire swarm of ultra-cheap robotic planes, rather than depending on expensive radar or lasers to locate and coordinate the flight of swarms or UAVs.

Created in the Laboratory of Intelligent Systems at the Ecole Polytechnique Federale de Lausanne (EPFL), the Swarming Micro Air Vehicle Network (SmavNet) project uses small (32-inch wingspan), lightweight (under 1 pound) UAVs with an electric motor and two control surfaces (ailerons and elevators) running on a single lithium-polymer battery with a flight-time of 30 minutes. The UAVs use GPS for location and WiFi for communications plus only three inexpensive sensors—a single MEMS gyroscope and two pressure sensors.

The swarm-interface requires only the simplest directions from a single operator on a ground-based computer, where algorithms send the control signals to each UAV—consisting of altitude, airspeed and turning rates. Currently the team is experimenting with two control algorithms. One is derived from the observation of ants, who use pheromone to coordinate swarms. The second uses machine learning to evolve unique algorithms for specific tasks.

The ant-derived approach uses airborne pheromones to set up a grid of responsibility for each UAV which patrols its sector while flying a circular pattern. The second approach uses unique flight patterns discovered by a machine learning algorithm, which are then reverse-engineered into a controller. Reverse-engineered controllers have so far demonstrated several useful behaviors not exhibited by the ant-derived algorithm, including exploration, synchronization and relayed communications.

Still in the development stage, the safety issues of controlling swarms of UAVs has been demonstrated to the satisfaction of the Swiss Federal Office for Civil Aviation, from whom official authorization for beyond-line-of-sight swarm operation has been granted for the tests.