Algorithm keeps autonomous vehicles working in the rain

Aurrigo International has developed a new rainfall algorithm that will help it detect the difference between drops of rain and objects whilst operating in up to 50mm of precipitation per hour.

Alongside new housing to better protect LiDAR sensors, the software breakthrough means the firm’s Auto-DollyTug can move baggage and cargo around airports in intense conditions that previously would have prevented manual operation.

This has been successfully tested in scheduled simulated rain and opportunistic heavy rainfall, both proved by observation and by rain meter readings.

“Traditionally, very heavy rain has presented a significant problem for autonomous vehicles, particularly for LiDAR detection and navigation,” said Simon Brewerton, CTO at Aurrigo International. “AVs using this technology rely on the reflection feedback of laser beams for localisation and object recognition and, in intense weather, the scattering and absorption of laser beams by raindrops can lead to distorted signals, compromising the vehicle's ability to accurately perceive its surroundings.”

He continued: “The first stage of the solution was to design the latest Auto-DollyTug with better casing protection for the LiDARs, which we duly did. However, the big ‘Eureka’ moment for us and the sector is the algorithm we have trained to strike a balance between the removal of raindrops and retaining the ability to detect real obstacles.

“This means that airlines and airport operators have complete confidence that, even in extreme rain conditions, our autonomous dollies will operate efficiently.”

Aurrigo’s technical team has ensured that the filtering of the rain is split into five spatial zones around the vehicle, with the strength of the filtering varying based on proximity.

Due to differences in the scan properties between the unfiltered and filtered scans, low-lying obstacles and reflective surfaces were not naturally detected and this had to be built into the software, leading to the adjustment of field height settings and ‘near field’ low profiles being added.

Sam Layton, regional VP Singapore at Aurrigo International, said: “The tests we did at a leading international airport had to prove the Auto-DollyTug could operate autonomously with no ghost or intermittent stops, react to a box and person injected into the vehicle’s path and safely navigate a T-junction airside with oncoming traffic left and right.

“All of this had to be achieved under 15mm of rainfall per the customer's requirement, which we successfully passed. Since then, we have increased our capabilities and operated airside on live trials with rainfall of 50mm per hour.”