Tiny drone with brain-like chip learns on the fly
HRL Labs, working under a DARPA program, gives early flight to artificial intelligence.
A Pentagon-funded program to develop computer chips with brain-like functions is bearing fruit on several fronts, most recently in the form of a neuromorphic chip that allows a tiny drone to learn as it flies around a room, thinking and acting on its own.
HRL Laboratories’ Center for Neural and Emergent Systems, with funding from the Defense Advanced Research Projects Agency, developed a prototype chip with 576 silicon “neurons” that was added to a drone aircraft weighing less than 3.3 ounces, according to MIT Technology Review.
In a demonstration in which it flew about in three rooms, the aircraft was able to process data from its optical, ultrasound and infrared sensors and recognize when it was in a new or a familiar room. HRL researchers said the chip’s neurons showed new electrical patterns in each room, in a rudimentary approximation of the way the human brain works.
The prototype grew out of DARPA’s Systems of Neuromorphic Adaptive Plastic Scalable Electronics, or SyNAPSE, program, which was launched in 2008 to develop neuromorphic, or brain-like, chips that could accomplish on their own tasks such as drawing information from video feeds and other sensor data, and provide decision support. HRL, IBM and Hewlett Packard received the initial grants.
In August, IBM unveiled a neuromorphic chip with 5.4 billion transistors designed to perform tasks such as pattern recognition, audio processing and motion control. And, like HRL’s chip, it does its work without drawing much power.
The SyNAPSE program puts a lot of emphasis on size, weight and power requirements, in addition to developing neuron-like interactions within a chip. The brain functions on far less power than a computer, so mimicking the brain should follow suit. IBM’s chip, with 1 million “neurons” and 256 million “synapses,” requires only 100 milliwatts of power, which DARPA said was an energy savings of two orders of magnitude over current chip technology. HRL’s 576-neuron chip, which took up just 18 grams of the test drone’s 93 grams (a little less than 3.3 ounces), required only 50 milliwatts in its test.
DARPA envisioned SyNAPSE as a way to develop chips that combine memory and processing in the synapses between neurons and do with on as little power as possible, the way a brain works. The goal is to develop systems that can assist warfighters in difficult situations, though the technology would have a lot of other, commercials applications as well.
The chips so far developed may still be a long way from practical, everyday use, but they could push the idea of autonomous systems to a new level.