End effector frame in the provided urdf for the Jaco7 does not match the urdf(xacro) from Kinova

[JCampbell9]

I'm working with the Jaco7 and noticed that the end effector frame described in the urdf provided does not match the one you get from kinova's ROS repo https://github.com/Kinovarobotics/kinova-ros/blob/3078ec6f00225e0ec5b80aae800bd3d94a59c369/kinova_description/urdf/j2n7s300.xacro

The one obtained from the relaced_ik_core doesn't include the pi/2 rotation for Yaw on the end effector. (first image is of the relaxed_ik and the second is the one from Kinova)

My question is if I change this value so it matches kinova's xacro/urdf will that have a negative effect on IK?
Will I have to retrain the neural network?

[djrakita]

Good question. This depends on whether or not the changed values in the urdf would change the resulting robot pose given the same robot state vector. For example, if joint values [0,0,0,0,0,0] had the robot going straight up with the old urdf values and now [0,0,0,0,0,0] has the robot facing forward in a different pose, the neural network would have to be retrained. In this case, it looks like you're only shifting the end effector frame, so my guess is that retraining would not be necessary. Feel free to give it a try, and if it seems like it is leading to errors, feel free to follow up here. Thanks