Hybrid application-specific integrated circuit has the potential to increase the operating time of intelligent swarm robots from minutes to hours.
When it comes to swarm robotics, the gap between science fiction and current technology is closing rapidly. A hybrid digital-analog application-specific integrated circuit (ASIC) is opening up new possibilities for intelligent, collaborative robot swarms by moving the burden of power consumption from processing centers to motor systems.
The new chip makes use of time domain computing, which encodes information in the width of a pulse that is carried over two different voltages. Time domain computing has gained traction in recent years in its capacity to increase the efficiency of image recognition processors while also decreasing the power required for neural network computing tasks. This most recent application combines both sensory recognition and reinforcement learning in a highly efficient 65-nanometer chip that’s designed to control palm-sized robots.
Above, Georgia Tech researchers provide proof of concept. The robots shown here feed information from sensors into the hybrid ASIC chip, which transmits instructions to a Raspberry Pi controller.
While the resulting hybrid system is slower than a purely digital or analog chip, it is still capable of learning from its surroundings. Robots enter their environments with a weighted neural network that allows them to recognize familiar structures while they explore new terrain. They bump into obstacles (including one another) as they learn, through trial and error, the general layout of their new world.
“We are sacrificing a little performance to get extreme power efficiencies,” said Arijit Raychowdhury, associate professor in the Georgia Tech School of Electrical and Computer Engineering.
At the International Solid-State Circuits Conference, Georgia Tech researchers demonstrated two robots that could navigate through a rubber cone obstacle course. That they avoided bumping into one another while negotiating unfamiliar terrain indicates a potential for mutual awareness. If collaboration can be defined, on a basic level, as learning not to bump into one another while circling plastic cones, then these robots have achieved it.
The ultimate goal of the project is to produce lightweight, power-efficient robots that are capable of understanding their own tasks as well as the roles of other robots. These machines would be used for deployment on reconnaissance and search-and-rescue missions as intelligent swarms, where they would require much less energy than a single robot but operate with (collectively) comparable computing power.
SRI International has been using microscale swarm robotics to build macroscale structures since 2014, but robots bigger than a bacterium still have trouble staying smart and mobile at the same time. If camera drones are going to cooperate with one another sans human input (the world’s first AI Instagram influencer, anyone?), they need to optimize power consumption in their motors as well as their processors. The team that developed the hybrid ASIC chip is working with micro-electromechanical systems (MEMS) researchers to decrease the power consumed by the robots’ motors and bring swarm robotics onto a visible scale.
For more on robot swarms, check out a team of intelligent robots that successfully printed a concrete structure and another one that can greet you with a smile.