It should be quite straight forward. The equations he uses are 'made up', and not based on anything really accurate. They just feel right. I adapted his stuff for tank steering (can't strafe), and I just had to come up with a decent approximation.

for the equations with minimum velocities, I guess you'll have to experiment. It's a very open system.

A simple guess,

velocity = truncate (velocity + acceleration, max_speed)

replaced by

velocity = truncate (velocity + acceleration, min_speed, max_speed)

but limiting with minimum velocities might cause problems (collision responses, ect..).

That's my understanding. It's a relatively accurate way of doing things, but for the seek pattern, it seems not very physiologically accurate. For the reach pattern, that's more like it.

As for moving a plane towards a aim point, it's like steering a car towards a similar aim point (say, an apex).

That will depends on the physical properties of the vehicle. For a car driven by an agent, you could come up with throttle and steering inputs given the car velocities, grip of the car, turning radius and braking force / engine force required to reach the target (that will generate centrigugal force), so that you drive on the limit. For planes, you can do the same, but the constraints are different.

His agent code is mostly for very basic pedestrians, who have little constaints (they can move quite hapily sideways, abd backward), and are limited only by max acceleration, and max velocities.

That's where his movement equations came from.

For my tank, I adapted that slightly because a tank can't strafe, and that gave me very simple tank A.I. good enough for games.

Anyway, whatever the type of agents, what you require from the agent, is his position, velocity, and the behaviour system will give you attract points, a repulse(?) point, that you pass to the vehicle so that the vehicle constraints can use that to generate new dynamics all by itself.

That's what I use anyway. A set of attractors and repulsors point / collision planes / whetever, passed to the vehicle by the A.I. behaviour controllers, and the A.I. driver use that, with the vehicle dynamics, to steer the vehicle away from danger, and towards the goals.

Pedestrians would do like in the docs, tanks would do it slightly differently, planes would do it another way. Depending on the distance, influence, and attraction of the control points / collision planes / ect..., and the vehicle limits, you steer, brake, accelerate, ect...

That's just my way of thinking. His system is great, but I don't think it's ment to be taken litteraly for more complex vehicles.

VEHICLE -----(position, speed, direction)---> BEHAVIOUR ---(attract point / repulse point / collision planes)---> DRIVER ---(steering / throttle / brakes)---> VEHICLE.