Science

Simple automotive collision-avoidance sensor inspired by insect brains

The bioinspired sensor gauges the distance of approaching cars based on the intensity of their headlights
The bioinspired sensor gauges the distance of approaching cars based on the intensity of their headlights

Self-driving cars typically use radar or LiDAR technology to avoid collisions with other vehicles. Scientists have now created a much simpler insect-inspired system that could serve the same purpose more efficiently – at night, that is.

While radar, LiDAR and computer vision systems are all reasonably effective at keeping autonomous cars from crashing into things, the actual modules themselves can only be miniaturized to a certain extent. They also require a considerable amount of power, plus they just generally add to the complexity of the vehicle.

Seeking a smaller, simpler and more energy-efficient alternative, Pennsylvania State University's Assoc. Prof. Saptarshi Das and colleagues looked to the insect world. More specifically, they studied the neural circuits that keep insects such as locusts from colliding with objects – and from getting caught by predators – while in flight.

The resulting optoelectronic sensor incorporates eight photosensitive "memtransistors" which are made out of a layer of molybdenum disulfide, and laid out in the form of a circuit. It measures just 40 square micrometers, and uses a few hundred picojoules of energy. According to the university, this is tens of thousands times less than the amount required by conventional collision-avoidance sensors.

Used at night, the device gauges the relative distance of cars simply by measuring changes in the intensity of their headlights – the brighter the lights, the closer the car. When tested in real-life driving scenarios, the sensor was able to predict two-vehicle accidents two to three seconds before they happened. While that might not seem like much, it would likely be enough time for an autonomous driving system (or the driver themselves) to take corrective action.

And although the technology probably won't replace existing systems, the scientists do state, "We strongly believe that the proposed collision detectors can augment existing sensors necessary for ensuring autonomous vehicular safety."

The research is described in a paper that was recently published in the journal ACS Nano.

Source: American Chemical Society

  • Facebook
  • Twitter
  • Flipboard
  • LinkedIn
4 comments
Chase
As with most current generation accident avoidance systems, the only reaction they have at their disposal is to slam on the brakes rather than actually avoid the obstacle. False positives, and even true positives, then have a tendency to lead to getting rear-ended, ironically causing the accident the system was intended to avoid. That said, I suppose a system built to mimic the level of intelligence of an insect is already an improvement over those brains that have grown reliant on current-generation ADAS.
byrneheart
If it works by reading increasing light levels at night, perhaps the opposite could be used as well, in measuring the occlusion of general light levels by objects during the day? If the system is that much more efficient than current tech perhaps it could be developed into a cheap aftermarket product for much cheaper cars
Jinpa
Every EV, PHEV and simple hybrid, along with many ICE (combustion) cars, with Adaptive Cruise Control (ACC) uses one or several systems to do its work. Also, given that the average age of a car on the road in the U.S. is about 12 years, there is a lot of variation of brightness of headlights because of the number of kinds of headlights. And deterioration of the coverings of headlights cause changes in brightness. Cheap abrasives on worn headlight covers damages headlight focus, turning intended spotlights into dangerous flood lights. So I think the intensity/distance assumption is badly flawed. The statement about ACC slamming on brakes shows lack of experience with ACC.
Louis Vaughn
If this system tracks each light source separately, i.e not the aggregate; a given source would be self relative. This would be important in distinguishing critical threats and respective rate/time to impact, etc. This combined with vector analysis and real time parallel processing, could be interesting. Hmm?