Several media outlets, including Electronic Design, recently spotted an interesting patent Tesla filed in 2017. Called “Technologies for Vehicle Positioning,” it seemingly addresses one of the major hurdles future self-driving vehicles face: the accuracy of GPS. As anyone with a smartphone probably knows, GPS positioning can be flaky, especially when indoors, or in close proximity to large buildings or trees. Apparently, Tesla wants to use machine vision and vehicle-to-vehicle communication to “correct” those GPS shortcomings. Electronic Design notes that all of the names on the patent are, or were, part of Tesla’s Autopilot development team.



An embodiment includes a method for using a set of vehicles as a set of moving reference stations for map-relative localization, the method comprising: receiving, via a hardware server, a map pose and a raw positioning measurement from a first vehicular client that is moving and vision-localized to a map; and sending, via the hardware server, the map pose and the raw positioning measurement to a second vehicular client such that (a) the second vehicular client determines a first position vector relative to the first vehicular client based on the map pose and the raw positioning measurement, (b) the second vehicular client receives a second position vector relative to a third vehicular client from the third vehicular client, and (c) the second vehicular client updates a second map pose based on the first position vector and the second position vector, wherein the second map pose is of the second vehicular client.

Discussion

Source: [H]ardOCP – Tesla Patent Describes Camera Assisted GPS System