Geshi Auto News? Although GPS (Global Positioning System) is a widely used technology for geographical positioning, it can play a navigation role while driving, but it is not always accurate. Apple Map (Apple? Maps) and other map applications, for various reasons, sometimes show users the wrong location.
Such reasons include GPS signal interference caused by trees or mountains, entering underground or indoors, reflected signals from urban buildings, solar storms, and even extremely rare situations, such as radio interference or shielding. However, this kind of problem will be encountered not only by GPS, but also by other global navigation satellite systems (GNSS), such as GLONASS (Russian satellite navigation system), Galileo (Galileo) and Beidou (Beidou satellite navigation system).
According to foreign media reports, on February 13, local time, the US Patent and Trademark Office announced a patent application of Apple, which was called "Machine Learning Assisted Satellite Positioning" (Machine? Learning AIDS? Satellite based? Positioning). In short, it is a method of comparing and analyzing GPS data with data obtained by machine learning model.
The idea of this patent is that the device will receive the estimated position information according to the GNSS signal, and then obtain a set of parameters related to the estimated position. Then, a reference position close to the estimated position of the device is provided to help with the correction.
According to the estimated position, the reference position and a set of parameters of the equipment, a machine learning model can be generated. Subsequently, the machine learning model can be used to estimate the specific location of the device reading the GPS number in the future, until a period of time passes, or the device is moved to a place where the parameters and model are inaccurate.
In fact, the device will use two sets of positioning data to generate a model to determine the distance between the GPS coordinates it receives and the actual position. For example, in a city with many high-rise buildings, the model can receive the information that the signal is reflected, and according to the wrong data, combined with the previous location reading and the general direction of traffic, get more accurate location information.
In addition, Apple added a requirement to consider the usage of the second device, including providing a model for others to use for storage. Apple also suggested using a Kalman filter (Kalman? Filter) estimates the data based on the noise measurement set, and considers "a lot of uncertainty" in the measurement and subsequent positioning to remind users that the position has changed, so that users can reconsider or ignore GPS data.
In recent years, Apple has been keen to increase its efforts in machine learning. 20 19 also hired senior Google AI scientists and AI experts? Ian. Goodfellow also acquired companies such as Drive.ai and Laserlike. A large number of public-oriented ML (machine learning) products of Apple are equipped with Siri, thus improving some location-aware functions.
From 2065438 to August 2008, Apple introduced in detail the use of geo-language model to increase Siri's understanding of local terms and locations, which helped to reduce the search based on points of interest by 18.7%. (The pictures in the text are all from appleinsider.com)
This article comes from car home, the author of the car manufacturer, and does not represent car home's position.