Kinematics interface with KDL implementation

[pac48]

Add kinematics interface with a KDL implementation

The proposed interface defines 5 public methods: initialize, update_robot_state, convert_cartesian_deltas_to_joint_deltas, calculate_segment_transform, and convert_joint_deltas_to_cartesian_deltas. These methods are sufficient to do basic end effector Cartesian control, but more methods would be needed if we wanted to also do forward and inverse dynamics.

The KDL implementation assumes a URDF has been loaded to the robot_description parameter.

One potential issue is that Eigen is allocating memory for its operations.

I would appreciate feedback anyone has (design pattern and naming conventions)

[dignakov]

I didn't say anything at the meeting, but I think this package is a great addition.

I do have one question/comment. There is already a place where the joint states are stored. What would be the reason for keeping a second set of state inside the kinematic interface? Could it not be a bit more functional where the joint positions, velocities, etc are passed into the functions?

[pac48]

I didn't say anything at the meeting, but I think this package is a great addition.

I do have one question/comment. There is already a place where the joint states are stored. What would be the reason for keeping a second set of state inside the kinematic interface? Could it not be a bit more functional where the joint positions, velocities, etc are passed into the functions?

That is a good point. If the user is require to pass the robot state in all of the other functions, (convert_cartesian_deltas_to_joint_deltas, calculate_segment_transform, and convert_joint_deltas_to_cartesian_deltas), then update_robot_state can be removed completely and the interface is almost stateless. The interface can still be in two state of course: initialized/uninitialized.

I guess the only real benefit to keeping the robot state internally is that the datatypes are different for each plugin. The values come in as vectors of doubles, but will need be copied to whatever type the kinematics interface uses. I am not sure that really matters that much.

[pac48]

I updated the interface to be stateless. Now each method requires you pass the joint positions in.

[bmagyar]

more to come tomorrow morning