Mike Horton, HYFIX.AI and Jens Windau, Lyft

View Abstract Sign in for premium content

Abstract:

Urban mobility application based services such as eBikes and eScooters are a growing environmentally friendly alternative to passenger cars, ride shares, and traditional taxis. However, cities have also had to deal with new challenges such as cluttering of the environment due to lost or misplaced mobility devices, as well as users riding in appropriate areas such as sidewalks, parks, or other common areas. Many of these inconveniences can be effectively mitigated by accurate meter-level localization of the rider; however, the accurate localization of an eScooter or eBike is particularly challenging. Deep urban canyon effects, the rider's self-blocking of sky-view, and the small antennae footprint are a few of the major GNSS challenges that make meter-level localization very difficult. In fact, this paper will show that standalone GNSS solutions on eBikes and eScooters in downtown urban environments can see larger than one hundred meter of error in some common conditions. 2-Wheeled platforms such as eScooters also experience significant dynamics and vibrations making the implementation of dead-reckoning complex. In this paper, a positioning solution that combines a low-cost, low-power, dual-band GNSS receiver (Broadcom's BCM4775X series) combined with sensor-based dead reckoning and optional GNSS corrections is able to reliably achieve more than a ten-fold improvement in localization accuracy over a standalone GNSS solution. The paper presents CDF data collected in downtown San Francisco during a series of trials using a consumer-grade IMU, wheel tick measurements, and GNSS corrections on a popular eScooter platform. The algorithm is shown to run on the BCM4775X series receiver, and also can positively impact power consumption. The paper also presents solutions initialization to handle rapid turn-on and convergence in these difficult conditions. Finally these solutions are also shown to be affordable and simple to integrate.