IEEE has published an interesting overview about Google's self-driving cars.

The advances in this field should be of interest to our ArduRover community. The article mentions that Nevada has updated its laws and has become the first US state to allow for these vehicles to legally share the road. No details about the process to "license" the "driver" is mentioned, but I'm certain that interested parties could find out more. 

 

The article talks about some of the challenges unique to this area of robotics. Specific behavior was coded into their 'autopilot' to communicate intent through action. The example of drivers at a four-way stop is used, where other drivers are not strictly following the rules of the road, and no driver moves, the car will first "announce" its intent to cross the intersection by moving more visibly, just like a human driver in the same situation. This is a domain-specific layer of complexity beyond that required for general robotics, related as it is to interaction with unpredictable agents sharing the robot's space.... such as humans, animals, and moving but "inanimate" objects.

 

This highlight a possible problem area in the design of our APM projects based primarily on DCM and PID controls. Our APMs have little "understanding" or modeling of the world, but for two-dimensionally-locked robots (well, we usually *hope* our ground and surface vehicles remain locked on the two-dimension plane... it usually means bad things when they leave it...) many which are successful have at least a basic method of modeling the world, its obstacles, and the physics involved. Beyond this, and outside of controlled environments, the robots also need to predict future conditions and the effect of its actions, and refold that information into the ongoing sense and avoid process, comparing and adjusting the whole time.

 

So where do we go from here, with that knowledge? And is there, perhaps, merit, to adding a "physics sensor", perhaps using a modern graphics chip created for the gaming industry? These are challenges which, perhaps, may take the "ArduSurface" projects in vastly different directions, but I suspect much of that work would subsequently be valuable to their flying counterparts. 

 

 

 

 

Views: 490

Comment by Curt Olson on October 19, 2011 at 9:41am

Hah, I think I saw one of these driving on the sidewalk yesterday. :-)


Moderator
Comment by Michael Pursifull on October 19, 2011 at 10:25am

oh. I don't think they belong on sidewalks ;)

Comment by arashi on October 19, 2011 at 11:58am

I've been tempted to attach the APM on top of my car and park in one of the "Autonomous Vehicles Only" parking spaces at the Nevada DMV Carson City.

Comment by Jack Crossfire on October 19, 2011 at 6:12pm

The wheel encoder is hilarious.  They really didn't have enough left over in the adsense budget for a speedometer tap?

Comment by Simon Wood on October 19, 2011 at 6:36pm

The wheel encoder is pretty standard for IMU/GPS navigation, however you can tap into the CAN-BUS to get pretty much the same information from the ABS sensors.

 

I wonder how 'Shelly' is getting along these days?

Comment by Curt Olson on October 20, 2011 at 6:08am

One thing about cars that the airplane guys might not think of is that cars have to drive under trees, under bridges, between tall buildings, inside parking garages.  Cars don't always have the ideal clear view of the sky that airplanes enjoy.  What do you do when you are left without gps?  Your mems gyros are noisy and might project forward a second or three before your errors begin to accumulate so much that it makes you unsure of where you are in your lane.  You can track forward motion + steering angle to help dead reckon, but you have no way to correct your errors without gps, and you don't know if there is any slippage or skidding (what if you were out driving in a recent snow-fall or icy conditions?)

 

There are some really interesting challenges with full size autonomous vehicles because your lane and safety margins are so slim, and you don't have good gps all the time.  Imagine an inner city freeway section where there could be a bridge every block or two you drive under -- or places like chicago where there is more bridge than sky above you in some areas near down town.  Places where you might lose gps every 10-13 seconds for minutes at a time at normal freeway speeds.

 

And you may be thinking that "my gps doesn't drop out in those situations". And I would suggest that maybe consumer gps's do a lot of cheating and estimating to make sure they don't appear to be dropping out, but all the fudging is hidden deep down so you don't know it's happening other than your track gets less accurate.

Comment by iangl on October 20, 2011 at 6:20am

Comment

You need to be a member of DIY Drones to add comments!

Join DIY Drones

Groups

Season Two of the Trust Time Trial (T3) Contest 
A list of all T3 contests is here. The current round, the Vertical Horizontal one, is here

© 2018   Created by Chris Anderson.   Powered by

Badges  |  Report an Issue  |  Terms of Service