I like the top (first) video the most... more than those agents. I think it looks cooler and more fun, it's more abstract, the agents seem more like an attempt at realism that isn't quite finished yet. But you said they weren't finished, so it's fine. their movement is just like... like I'd never move
again, very nice work
~"I never let school interfere with my education"~ -Mark Twain
The both methods will be combined. Think about a basic node related pathfinding plus the obstacle avoidance when obstacles accour.
I'm still working on the obstacle avoidance. Also i have to find a solution for "Controlling Agents in High-Density Crowd Simulation"
Video not from me:
PS: to behave the individual "more human" the trick is to set more waypoints and let the individual direct to the next visible node with the shortest way to the destination. This way the individual will walk on shortcuts - like a human does.
It would be very interresting to make a simulation of cars in the city or on a high-way. By djusting the behaviour of the cars you can show why traffic jams happen, and prove/disprove that speeding or slow driving is the cause.
Also the layout of the streets can be planned this way.