tud-airlab / mppi-isaac

Model Predictive Path Integral Control using isaacgym for rollouts, gpu-accelerated
166 stars 19 forks source link

Bug: point robot goes through obstacles #44

Open cpezzato opened 1 year ago

cpezzato commented 1 year ago

In the pybullet example of point_robot_with_obstacle.py, the point robot does not collide with the obstacles in the rollouts, and can instead go through. This does not happen for the panda robot though.

Checking in the Isaac gym wrapper, the obstacles have collision=true, @c-salmi where could this bug come from then? Also in the point_robot.urdf collision meshes seem okay.

c-salmi commented 1 year ago

I think it's a similar bug to #39 where the last actor of the environment is not considered for collision checking. I'm still not exactly sure where this comes from, since, as you pointed out, the collision attribute is set to true.

It should be possible to work around it though by adding an actor that's not important, as the last actor in the environment. However, for the pybullet examples this might be more tricky, since the actors are dynamically added (and the simulator is reloaded) when the observation has an obstacle that's not in the world. So the obstacles will always be added last.

I'll think about a nice way to fix it for the pybullet cases.