@newuhe I notice two things from this script
- See the lines
bad_proportional_gains = articulation_controller.get_gains()[0]/50
articulation_controller.set_gains(kps = bad_proportional_gains)
These are intentionally setting the PD gains to incorrect values to facilitate the usage of RmpFlow Debugging Features. You should copy from a different part of the RmpFlow tutorial. Try here: RmpFlow Tutorial Code Snippet
- In the Code Snippet I directed you to, you’ll see that RmpFlow is initialized like so
rmpflow = RmpFlow(
robot_description_path = rmp_config_dir + "/franka/rmpflow/robot_descriptor.yaml",
urdf_path = rmp_config_dir + "/franka/lula_franka_gen.urdf",
rmpflow_config_path = rmp_config_dir + "/franka/rmpflow/franka_rmpflow_common.yaml",
end_effector_frame_name = "right_gripper",
evaluations_per_frame = 5
)
Notice the line: end_effector_frame_name="right_gripper"
The RmpFlow algorithm uses the Franka’s URDF file as its source of truth, rather than USD. You can set any valid frame in the URDF to be the end effector that RmpFlow should use. The “right_gripper” frame is a frame that we added to the Franka URDF that does not exist on the USD stage. As you have encountered, this can be a little misleading. The line my_franka.get_end_effector is returning the USD Prim at the path “panda/panda_rightfinger”, which does not exactly match the frame that RmpFlow is using for position control. To get the end effector position that RmpFlow is using, you should use the line:
rmpflow_end_effector_prim = rmpflow.get_end_effector_as_prim()
The rmpflow_end_effector_prim will then track the position of the end effector over time, which you can query with rmpflow_end_effector_prim.get_world_pose().
One of our goals for the near future is to use USD as the single source of ground truth, which will eliminate this form of inconsistency, but for now, RmpFlow (like other common implementations of manipulation planning and control algorithms) relies on a URDF file.