Thread by @jpr007

@mgereb @ajtourville So let me start by saying that my frame of reference is the human driver

And we need to respect three things about this human driver :

@mgereb @ajtourville 1. We see that human driver moving around in time T within a 3D physical space with fixed reference points like road, lamp posts, buildings

- so the classic visualization of this is like a time-based video game with the vehicle shown inside a 3D map of its surroundings

@mgereb @ajtourville 2. But actually the driver is in an Einstein relativistic space where he sees everything relative to his own continuously moving position

- he is the only fixed object and everything else is moving relative to him with complex vectors

@mgereb @ajtourville 3. In addition to instantly analyzing everything around him in relation to this relativistic space, this human driver has an extraordinary ability to anticipate changes in the movements of other objects around him

- whether they be a vehicle, a human, or even a cat

@mgereb @ajtourville So the first three dimensions that this human driver is able to instantly observe and calibrate for shape, size and distance are defined by our classic x, y, z

- but in perspective form and in relation to his own position before relative to any other object

@mgereb @ajtourville The next three dimensions represent the velocity vectors of every object, including their rotational and tumbling vectors

- this is more complex than it sounds because he and they are moving relative to each other as well as relative to their own position and orientation

@mgereb @ajtourville The third set of dimensions are the changes in all of these velocities, whether it be accelerations or decelerations in any of them

- we may think of those changes relative to the object itself but the driver is actually seeing and assessing those changes relative to himself

@mgereb @ajtourville So now we have 9D

And some may argue that these 9D can be approximated by 3D + T for terrestrial driving purposes

- and they may be right, depending on how the internal algorithms handle the available data

@mgereb @ajtourville But then we must remember that any human driver - or walker for that matter - has a further 3Ds of ability to pick up on visual and audible cues that allow him to ANTICIPATE changes in these other 9Ds

- a tennis player provides an illustration of this

@mgereb @ajtourville And a vehicle driver uses this skill continuously, especially in slower moving environments filled with bicycles, people, motorbikes, traffic lights, complex road rules and other factors that are not simple to capture or express in the first 9Ds

@mgereb @ajtourville And some of these anticipatory dimensions can be “slow burn” like the behaviour of vehicles approaching a left-turn signal

- or they can require “instant” response like a child darting out into the road in pursuit of a rolling ball

@mgereb @ajtourville We may argue that current systems are trying to capture and recognize all of these 12D even while working within a compressed 3D + T logic environment

@mgereb @ajtourville And perhaps they will find ways do it even though they must inevitably have less rich visualizations and cognizance than the human driver

@mgereb @ajtourville One of the most interesting implications of all of this actually comes from the initial point 2.

“the driver exists in an Einstein relativistic space where he sees everything relative to his own continuously moving position”

@mgereb @ajtourville This highlights the low utility of attempting to prepare highly detailed mapping in advance of a vehicle’s arrival in a designated area

@mgereb @ajtourville Low quality mapping is helpful for general navigation

- showing where roads lead to, and how to enter and leave highways, and where to expect intersections, speed control zones, schools, etc

@mgereb @ajtourville But high quality mapping is largely useless even as soon as it is created because it will never be portraying the actual view from the driver’s Einstein space

- it will first have to go through a complex transformation process

@mgereb @ajtourville - then it will have numerous weakness relative to simply observing the view outside the vehicle

- and it will be carrying lots of extraneous and unnecessary data which will give the impression of precision but in fact will only be serving as distraction

@mgereb @ajtourville So in practice it should prove to be more valuable for mapmakers to focus on reliable and frequently update general mapping with high quality navigation protocols

- than trying to create highly detailed maps that have little or no intelligent uses

@mgereb @ajtourville @threadreaderapp unroll