Tuesday, December 04, 2012

Distributed Processing and Robot Taxis

I went to a seminar earlier today that was concerned with getting networked dynamic systems to pursue their own goals and then merge on a common state at the end of a certain time interval.  Pretty abstract, and I have to admit that the math was impenetrable to me.  But it got me thinking - what if the systems went to a distribution, rather than a single point?  And, what if the times were arbitrary and mismatched?

I really, really want my next car to be able to self-drive.  It seems feasible, too, if I can hold out for a few years.  One of the things that commentators bring up with self-driving cars is that the model of individual car ownership may lose a lot of its advantage.  If cars can drive themselves, why should a car have to wait for its owner to get back?

In New York, people take taxis under certain circumstances.  Taxi drivers know the times and places when people want taxis the most, and try to put themselves there.  This works well for knowledgeable taxi drivers because they can stay quite busy, and works relatively well for potential customers, as long as they follow a common pattern.  It's not so great for someone doing something a little unusual, though, the taxi fleet has a limited ability to scale, and the way that taxis are leased by drivers often means that the taxi fleet is operating well below the capacity of the equipment.  Drivers get tired, and don't use all 24 hours.  (Nor would I want them to!)

Imagine if a city had a fleet of self-driving taxis that networked with one another and reported their states to a central dispatcher.  Further, imagine that the dispatcher could receive calls via cell phone or app.  The communication was a key aspect of the seminar today.

The selfish vs. team aspect of the individual systems' behavior was a big part of the seminar's theme.  But instead of having synchronized behavior changes, the taxis would decide based on different weights assigned to their needs at different times of day.  During commute hours, all reliably operable taxis would stay available until they picked up fares.  During lower-traffic periods, fuel and maintenance needs would be assigned greater importance.

The team aspect of the taxis' behavior would be to try to maintain a certain geographic distribution, with varying densities based on frequency of calls.  A lot of taxis would attempt to travel to areas with historically greater call frequency when available, while fewer taxis would station-keep in areas with low demand.  By networking with nearby taxis, taxis could put themselves in the right places to maintain the right gradient in available taxi density.  Having all taxis maintain good records and download information to the same place would facilitate development of really good statistics about when and where the demand was.  Taxis would know to go to residential areas in the mornings, to bring people into commercial cores, they'd know to be around offices for evening rush hour, they'd know when the bars close.  Because the taxis would be trying to maintain a distribution as a team, this behavior wouldn't imply abandoning a neighborhood - a few taxis would still stay in areas with lower call frequency in order to maintain the right proportional distribution of taxis, so someone using a vehicle against the general pattern wouldn't have a taxi too far away.

Without human drivers, robo-taxis could be operative during almost all hours, with a greater proportion refueling and undergoing maintenance at off times.  They could potentially serve many more fares in a day, and without a human driver, they could be cheaper.  Maybe a fleet of robo-taxis would fit into the right need to take a few cars out of parking lots and spaces, maybe off the roads entirely - they'd reduce people's need to drive to work "just in case" or because they'd be carrying something heavy for a small part of their return trip.


WhatHouse.co.uk said...

Think we are a long way off self drive cars and robot taxis! Interesting concept though!

Station Cars

Andrew Wright said...

I give it about ten years. Maybe five. They're already street-legal in a couple states in the US, so all it needs is someone to put the technology into a commercial package.