# Dependent driver

Hi,

Is there a way to set a driver which depends on other drivers states?
I mean something like adding a listener to a driver which captures the state of another driver and act accordingly?

Best,
Aff

Hi Aff,

I am sorry this is not currently possible, it would indeed be a nice feature to have.

Could you describe what this is needed for?, there might be a different way to do this

We have ways to formulated conditional kinematics, to avoid e.g. a foot penetrating a floor.

Best regards
Søren

Hi Søren,

Thanks. Well, I am thinking about using Anybody to generate an environment for testing reinforcement learning algorithms. imagine a cart-pole example which is to control an inverted pendulum by controlling the velocity of the cart carrying the pole.
The hinge joint between the pole and cart should be controlled by the horizontal movements of the prismatic joint between the cart and the global reference frame.
I have also been in contact with John Rasmussen but he mentioned that there is no straightforward way of doing it but there may be some heuristic ways for doing it. so I posted this message here to see if anyone else has thought of something along that line.

Best

Hi Aff,

If the model is only for testing i assume that you have an estimate of the motion. Then you could apply the motion of cart and pole. If you then have an artificial torque present at the hinge joint, the motion will be correct when this torque is close enough to zero, this would be a simple way to test the motion.

Does it make sense?

Best regards
Søren

Hi,

Well, the idea is to predict the motion. We do not have the kinematics of the revolute joint between the cart and pole but based on the bio-mechanical structure of the model, we know it is going to be a function of the behavior of the prismatic joint between the cart and GRF.

so I think the only possible way of handling it is to find a way of implementing a dependent driver in one way or another.

BR

Hi Aff,

If you want AnyBody to predict this motion it could be done through optimization.

The objective would be to minimize the artificial torque added, the design variables would be angle and angular speed. This is possible to do inside AnyBody using a gradient based optimization, but i imagined that your learning algorithm would have done the same.

Best regards
Søren

This topic was automatically closed 125 days after the last reply. New replies are no longer allowed.