Apple, which is considered to be a benchmark for electronic gadgets, has applied for patenting of a cutting-edge Machine Learning based Global Navigation Satellite System (GNSS) device.
Global Navigation Satellite System refers to a group of satellites, enveloping the globe, to aid in navigation. GNSS satellites triangulate user's location and send signals from space that transmit positioning and timing data to GNSS receivers, which determine a location by finding the best estimate after combining signals from various satellites. The available GNSS systems include Galileo of Europe, NAVSTAR Global Positioning System (GPS) of USA, Global'nayaNavigatsionnayaSputnikovaya Sistema (GLONASS) of Russia, BeiDou Navigation Satellite System of China, and NAVigation with Indian Constellation (NAVIC) of India.
At present, almost all the available GNSS systems use Kalmanlinear quadratic estimation filters. These navigational systems rely totally on the orbiting satellites, hence, inherently susceptible to inaccuracy due to such large distances between satellites and moving users on the ground. Now, Apple, as per the filed patent, seeks to use Machine Learning along with the Kalman filters to improve accuracy of the GNSS devices and bring it down to centimetres from metres.
According to the patent application filed, Apple has proposed Machine Learning Assisted Satellite Based Positioning:
- A device that implements a system for estimating device location based on a positioning system comprising a Global Navigation Satellite System (GNSS) satellite, and receives a set of parameters associated with the estimated position.
- The processor is further configured to apply the set of parameters and the estimated position to a machine learning model that has been trained on a position relative to the GNSS satellite.
- The estimated position and output of the machine learning model is then provided to a Kalman filter for more accurate location.
- The device, say an iPhone would generate a machine learning model, for example, by comparing GNSS position estimates (or estimated measurement errors) with corresponding reference position estimates (where the reference positions correspond to ground truth data).
- In one or more implementations, the ground truth data may be better (e.g., significantly better) than what a mobile device alone can perform in most non-aided mode(s) of operation. For example, a mobile phone in a car may be significantly better aided than a pedestrian device, because the motion model for a vehicle is more constrained, and has aiding data in the form of maps and sensor inputs.
Tall buildings and tree cover can confuse the positioning systems to accurately locate the user. So, Apple seeks to generate machine learning models on the device that would predict the user's location based on training as well as a reference position.
As of now, GPS is the most reliable and widely used satellite location system, but, Apple's application, as per filed patent, goes beyond it to include various types of GNSS systems, such that in each case triangulation of satellite data, the machine learning-corrected version will be transmitted to the Kalman filter. This would result in a single estimated device location at any time instant, which will be in-line with data from both the current user and the machine's training.
Currently, the GNSS systems generally rely on live location-determining through multiple satellites, which can consume some time, during which the user may move. But, such systems face various noises, such as dispersion of signals, reflection of signals, due to large distance between the satellites and the GNSS receiver, atmospheric envelope, etc. But, in case of the proposed Machine Learning based GNSS device, the device can be trained to provide improvised location of the user based on previously gathered data from the surroundings. In a particular locality, there might be a set pattern of satellite signal reflections, which can cause errors while determining user location, here the device can overcome the problem through machine learning.
Apple's patent application, which was officially published on 13 February 2020, could be added to the company's navigation software, if granted, or may remain in the labs indefinitely.
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.