Researchers have created an ultra-low power 2D transistor to mimic the collision-avoidance neurons of a locust in their autonomous robots. Scientists from the Indian Institute of Technology Bombay and King’s College London collaborated on the study to explore low-power solutions for autonomous robots and vehicles, which are growing in prominence.
Autonomous driving and motion have long been a holy grail for machine learning and AI developers and researchers, and collision avoidance is the key to making the tech feasible in the real world. To this end, the IITB and King’s College students set out with the goal of creating a collision solution on extremely low power.
In studying collision avoidance, scientists discovered a collision-detecting neuron in locusts. Called LGMD (lobula giant movement detector), this neuron spikes when large objects come near the locust, helping the insect avoid danger. This neuron was able to be duplicated by scientists with incredibly thin two-dimensional transistors, which also produce spikes analogous to the locust neuron and for a similar energy cost: less than 100 picojoules (for context, running a 100 W incandescent light bulb for one second costs 100 joules of energy). The thin and cheap transistor was also fully functional, being able to be reprogrammed to look for different types of movement and successfully avoid obstacles with high degrees of accuracy.
A 2D transistor is an impossible dream for large-scale chip manufacturers, as when transistors become smaller, they also become more energy-efficient. Of course, the transistor used in the IITB study is very simple, spiking when movement is detected within a range and nothing more. But the authors have a vision for where this two-dimensional tech can go after this study.
These super-efficient transistors could help greatly with the energy cost of the often-inefficient AI technologies we have available today. Professor Bipin Rajendran, at King’s College London and co-author of the study, writes “We demonstrated that this spiking neuron circuit can be used for obstacle detection. However, the circuit can be used in other neuromorphic (systems mimicking the human brain) applications based on analog or mixed signal technology that require a low-energy spiking neuron.”
If you’re curious about more details and the scientists behind the study, you can check out the study here. We’ve also written much about AI moving itself around places recently. Check out our piece about ChatGPT trying to play Red Dead Redemption 2, or perhaps about how China has used an Nvidia chip to hypersonic weapons for better autonomous flight.