Locust vision could help self-driving cars, drones avoid collisions

Scientists copy mechanism that the insects use to ensure crash-free flights.

K. S. Jayaraman

doi:10.1038/nindia.2020.129 Published online 28 August 2020

Locusts use insect vision to avoid bumping into each other.

© Unsplash

How do desert locusts, billions of them flying in swarms across the sky to attack crops, avoid crashing into one another? They deploy their ‘insect vision’, the unusual ability that helps these pests undertake long collision-free flights.

A team of Indian scientists at the Pennsylvania State University (PSU) in the US is now trying to mimic this unique ability to build collision detectors1, which can help robots, drones and even self-driving cars detect an upcoming collision and steer away from it.

"Insect vision offers an ideal model system for task-specific visual information processing circuits such as collision avoidance", says Saptarshi Das, assistant professor of engineering at PSU.

Das and his colleagues started looking at how this works in locusts. They found that locusts were able to avoid collisions by quickly changing directions using a single, specialized neuron, called the Lobula Giant Movement Detector (LGMD). The neuron receives two different signals: one, the image of an approaching locust (which gets larger as it comes closer) and the other, the angular velocity between itself and the approaching locust. This single neuronal cell expertly computes the changes in the two inputs to quickly figure an escape response when it senses an imminent collision.

The researchers built a nanoscale collision detector – a chip with an area of 2.25 mm X 2.25 mm – that mimics the escape response of the LGMD neuron in locusts. They made the compact detector from a layer of molybdenum disulfide (MoS2) photodetector stacked on top of a non-volatile and programmable floating gate memory architecture. The detectors right now being used in autonomous automobiles are very large and heavy. Their detector, the researchers say, is smart, low-cost, task-specific, energy efficient and miniaturised.

They have tested the device only with objects on a direct collision path and are yet to optimise its response in other practical situations. They are also examining if multiple devices on the same chip can help avoid collisions in a 3D space. 

The researchers claim this is the first demonstration of “in-memory computing and sensing” using a single nanoscale device. Emulating the LGMD neuron using a MoS2 photodetector is a new milestone for bio-mimetic devices, confirms Deblina Sarkar, the founding director of the Nano-Cybernetic Biotrek Research Lab at the Massachusetts Institute of Technology (MIT).  It is an interesting example of how understanding the computational aspects of information processing in the nervous system can inspire innovative solutions to real-life engineering challenges, according to Fabrizio Gabbiani from the neuroscience department of Rice University in Houston, Texas.

Das and colleagues are in the process of patenting the technology.


1. Jayachandran, D. et al. A low-power biomimetic collision detector based on in-memory molybdenum disulfide photodetector. Nat. Electronics  doi: 10.1038/s41928-020-00466-9