Maybe the most impressive bicycle control simulation I've seen

[Jason Moore]

Some graphics folks at Georgia Tech have used neural networks and genetic algorithms to teach a bicycle/rider dynamics model to balance and direct itself along with some very impressive bicycle stunts like endos and wheelies. See their work here:

http://www.cc.gatech.edu/~jtan34/project/learningBicycleStunts.html

As graphics and character animators typically do, they discount more detailed dynamics models, but they do have some sort of bicycle dynamics model that at least seems to behave realistically. It's likely that the same techniques could be applied on another model.

Jason
moorepants.info
+01 530-601-9791

[Alessandro Belli Design]

Hi Jason,

we completed a major step in the development of the bike.

There are already on the web a number of pages, try Google "Bike Intermodal"

Also our site: www.bike-intermodal.eu

Big, single, truck hugs,

Alessandro

--
You received this message because you are subscribed to the Google Groups "Single Track Vehicle Dynamics" group.
To unsubscribe from this group and stop receiving emails from it, send an email to stvdy+un...@googlegroups.com.
To post to this group, send email to st...@googlegroups.com.
Visit this group at http://groups.google.com/group/stvdy.
For more options, visit https://groups.google.com/d/optout.

[Damian Harty]

I've done this to some extent in a commercial multibody code. I've been modelling basic control above walking pace for quite a while using a PID controller on steer torque and with roll angle as a target. With relaxation in the tires, it takes a little more tuning but without it, it works very easily.

I also hooked in throttle and brake torques and I can set a front wheel spindle height target for a wheelie, rear wheel spindle target height for a stoppie (endo). 

What was much more interesting was that I was able to drift the motorcycle using a target rear wheel spin state with no real modifications to the basic controller, even to the point of lifting the front wheel crossed up. If you drop the spin state target mid drift, the bike high-sides and if you are too greedy based on the current lean angle, the bike low sides. It all played out very uncontroversially. Which is not to say I can do it in real life...

Damian