Kit. wrote:I don't see anything that I could recognize as "GPS mapping" there.
Can you give a definition?
Our notions of what navigation is obviously differ. That's why these criteria above could be useful to continue meaningful discussion.
There are two senses in which the word "mapping" is used here. What I'm referring to as "GPS mapping" is the construction of a representation of the external world as it appeared at some point in the past, which was constructed at some point in the past, independently of the use to which is would be presently put. Typically it would be used for navigation. The predominant form nowadays is a set of datapoints georeferenced to GPS coordinates, displayed on a screen which human drivers can refer to, or stored in a data structure which robotic drivers can refer to. An older form involved paper and ink drawings reproduced on a printing press and often accordion-folded to make them easy to store.
For this purpose, navigation is the act of making strategic plans and decisions as to which course to follow in order to arrive at a desired location. This typically will involve the choice of roads to follow, intersections to turn at, and sometimes time-and-distance calculations to help the driver decide when to depart. Navigation is typically pre-planned, but can also be updated on the fly when conditions warrant.
The other sense of "mapping" which is used here (but not by me) is the system by which an entity is aware of its actual surroundings, taking in sensor data, and processing it in order to figure out where things are around it. Sensor data is remembered, and the relative positions of objects in its field are recalculated as needed when the entity moves and has not updated its sensor data. People create mental maps of rooms they are in; they can use these mental maps to navigate for a limited time when the lights go out or they shut their eyes. Bats create a mental map of their surroundings based on echolocation. A "mental map" in this sense is largely another name for a kind of self-awareness.
Kit. wrote:I'm not talking about "acts", I'm talking about decisions. What are your criteria whether a decision belongs to the "navigation" system or to the "control" system, and what results are produced by applying these criteria to a decision to change lanes?
In short, control is a set of acts. Navigation is a set of decisions. Actions of course require decisions (how far to turn the wheel, when to start turning back...) but the primary focus is the action
of changing lanes without hitting anything. That is "control". It is a tactical thing.
Navigation is a strategic thing - it is the... er... act of planning
a route to a destination. I am not using the word to refer to dodging ("navigating around") an obstacle.
Another distinction is that navigation (as I'm using the term) uses outdated nonlocal information for planning purposes. The map is compiled beforehand, just like a paper map is. An out-of-date map could get you lost, but it shouldn't get you into an accident.
Control (as I'm using the term) uses up-to-date local information for collision avoidance. Sensors provide this information, which includes exactly where the curb is, how wide the lines are, where the nearby vehicles and other obstacles are, and the location (and existence!) of the pothole in front of you. Poor control isn't likely to get you lost, but it could easily get you killed.
There seems to be a widespread feeling that we need "better maps" (of the GPS or paper variety) in order for autonomous driving to be successful, and therefore the (i.e. google) mapping vehicles need to gather more info. I say no, the maps are fine. We need better sensor interpretation. Better "mental maps", as it were.
While I'm comfortable with my neighbor knowing when I ride my bike and what my route is, I am absolutly not
comfortable sharing that information with google, amazon, waymo, uber, lyft, Tesla, GM... They have no business knowing my life, and the argument "it will make the robots drive better" is not at all compelling.
Kit. wrote:Yeah, it's easier to blame technology for being "a bad thing" than to keep "bad people" from abusing it. The problem is, you cannot force your competitors (China, for example) to do the same, so, by not crippling their technological progress, they may get a competitive advantage over you. Google is already expanding their AI research into China, even though they still have troubles with providing their core services there.
Yes, and this is a bigger problem than the sliver of it I bring up here. And in any case, I'm not blaming technology for being a "bad thing", I am saying that self-driving cars come with technology that is easily used for bad things by powerful entities, and these powerful entities are trying (and succeeding) in convincing us that "these aren't the droids you're looking for". We are, for example, inviting 24-7 surveillance machines into our homes
, and telling them all our secrets. Echo, Alexa, Nest, all of them are "ooh, shiny" ways to give control of our data to the companies behind them; it's a land grab that's happening before we realize there's land to grab. Does nobody realize the implications of the disappearance of local sync? But that's another rant for another day.