How New Electric Scooters Use AI To Warn Of Pedestrians, Road Hazards

As electric scooter demand increases due to concerns about public transportation during the pandemic, companies are using AI to improve rider safety, reduce sidewalk congestion and manage parking. 

Installing next-gen GPS with AI provides both the rider and the scooter company identify potential obstacles like pedestrians, baby carriages, cars, trucks, buses and other scooter users which can also help municipalities better organize transportation. JL

Paul Sawers reports in Venture Beat:

Voi is to become the “first in the world” to deploy computer vision and on-device AI on scooters to help avert collisions and issue alerts if scooters are being used contrary to local requirements. It’s using real-time knowledge to help scooters understand their environment. (Its) next-gen GPS smarts pinpoint the scooters to within three centimeters, managing scooter parking and minimizing sidewalk congestion. Its computer vision-powered sensors enable scooters to detect pedestrians and determine whether a scooter is on a road, sidewalk, or cycle lane.
European electric scooter startup Voi is turning to artificial intelligence (AI) to enable its vehicles to detect pedestrians and sidewalks, a move designed to ingratiate itself to municipalities worried about chaotic micromobility schemes. 

Escooter (and ebike) sales have surged throughout 2020 as travelers sought alternatives to crowded public transport, but this proliferation serves to stoke existing safety concerns around how they’re used and deployed in busy cities. Voi is now looking to become the “first in the world” to deploy computer vision and on-device AI on scooters, to help avert collisions and issue alerts if scooters are being used contrary to local legal requirements. It’s ultimately about using real-time knowledge rather than guesswork to help scooters understand their environment. 

“Voi is developing scooters that can ‘see’ what’s around them and therefore irrefutably ‘know’ what they need to do in order to be safe, whereas other scooters are trying to ‘feel’ what’s around them and use that to ‘guess’ what they should do next,” Voi CEO Fredrik Hjelm told VentureBeat. 

Founded out of Sweden in 2018, Voi is one of a number of escooter startups that enable citizens to download a mobile app, find the nearest scooter, and pay for each minute that they need it. 

The company, which operates across dozens of cities in Europe, has raised more than $250 million in VC investment in the past couple of years and achieved profitability for the first time this year, bucking a trend that has blighted similar companies in the burgeoning micromobility space. 

Central to Voi’s model so far has been working hand-in-hand with local authorities as part of a partnership, a move that has seen it score exclusive pilot deals in the U.K., which fast-tracked escooter trials back in May in response to the COVID-19 crisis. One of these pilot markets is in Northamptonshire, where it will test real-time pedestrian detection through a partnership with Irish AI startup Luna. 

Founded out of Dublin in 2019, Luna emerged from Intel’s Edge AI incubator program last year, where it leveraged Intel’s Movidius Myriad X Vision chip to develop computer vision algorithms aimed squarely at scooters. Edge AI is where algorithms and the associated data are all stored and processed locally on a hardware device, rather than relying on a remote server to carry out the spadework — this ensures faster speeds due to lower latency and promotes a more privacy-centric ethos. 

Luna’s technology constitutes several components, one of which is next-gen GPS smarts that can pinpoint the scooters to within three centimeters, which is important in terms of managing scooter parking and minimizing sidewalk congestion. But arguably the more interesting technology is its computer vision-powered sensors that enable scooters to detect pedestrians and determine whether a scooter is on a road, sidewalk, or cycle lane, making it easier to enforce local riding restrictions. 

Voi isn’t the first escooter company to attempt this. Earlier this year, Lime announced plans to thwart sidewalk riding through “understanding” the vibration of the underlying surface. But Lime was only able to issue a warning afterward by crunching sensor data gleaned from the speedometer and accelerometer. 

A few months after this implementation, the City of San Jose, California issued a report noting that while Lime was the only operator to make significant headway in preventing sidewalk riders, its inability to detect the activity in real time was a major drawback. The council wrote: 

To date, no operator has developed a scalable and reliable technology solution to detect sidewalk riding in real time. One operator, Lime, has demonstrated a scalable partial solution, but Lime is not yet able to detect sidewalk riding in real time nor slow e-scooter speeds on sidewalks. Furthermore, the effect of Lime’s proposed approach on sidewalk riding and pedestrian safety is unknown at this time.This is something that Voi wants to address with its Luna partnership, using computer vision techniques to enable real-time surface detection and open the door to more useful tools to keep scooters away from places they’re not supposed to be. This is where its edge AI truly shows its worth, as it negates the need for huge bandwidth and compute capabilities: It can “control and govern the scooter” in real time. 

“The Luna and Voi solution uses a ‘camera as a sensor’ alongside lane segmentation AI, similar to what you would see on a high end car,” Luna cofounder and chief business officer Ronan Furlong told VentureBeat. “This gives us data which is much much closer to ‘ground truth,’ while the algorithms can be easily trained for the idiosyncrasies of each individual city.” 

Perhaps more importantly though, Voi and Luna are specifically targeting areas where scooters come into close proximity to humans, not only detecting pedestrians in real time but counting them to figure out how congested an area is and intervening. 

“Voi and Luna will be developing a series of alerts, notifications, and possible interventions, for the purposes of enhancing pedestrian and rider safety,” Furlong continued. “For example, the technology creates the opportunity for Voi to remotely slow down scooters whose cameras are detecting a certain number of pedestrians in their field-of-view and direction of travel.” 

This integration also means that Voi will have to tinker with its scooters to fit the appropriate camera sensors. The initial prototype unit is simply tethered to the top of the scooter, but Luna’s vision module will ultimately be integrated directly into Voi’s connected telematics unit, meaning that riders won’t notice any changes aside from a small camera lens. 

With demand for clean micromobility sky high, this will undoubtedly lead to heightened competition across the escooter sphere. Indeed, just this week, Voi rival Tier announced that it had raised $250 million in a round of funding led by Softbank’s Vision Fund. Tier also recently announced plans to introduce audible alerts to its scooters, designed to alert blind or partially sighted people of a scooter’s approach. 

While its initial focus will be on gathering data for pedestrian detection as part of the year-long pilot in Northampton, Voi’s technological integration with Luna will play a big part in its “upcoming tenders for major city markets” and could prove pivotal to winning them over. 

“We are seeing various scooter operators scrambling to announce technology add-ons which purport to solve the challenges that the industry is facing — particularly street clutter and sidewalk-riding,” Hjelm said. “However, we view these announcements as incremental improvements, rather than innovation that is genuinely moving the dial in respect of safety and governance of shared scooters.”