coriolis() and Robot.copy().rne() fail for URDF/ETS robots: copied links keep their original parent #730

robot.coriolis(q, qd) raises ValueError: Link(...) is not in list for every URDF-based Robot I tried: UR5, KinovaGen3, LBR, and a two-link URDF with full inertials. DH robots are unaffected.

coriolis calls nofriction(), which calls copy(). The same error occurs for robot.copy().rne(q, qd, qdd) directly, while rne on the original robot works. Cause: after copy() each link's parent (and the joint's parent) still refers to the original robot's link object, so self.links.index(joint.parent) in Robot.rne (group_of_link_idx.get(...)) fails.

Repro (any URDF with dynamics):

import numpy as np
from roboticstoolbox.models.URDF.URDFRobot import URDFRobot

r = URDFRobot("/path/to/two_links_robot.urdf")   # two revolute joints, with <inertial> elements
q, qd = np.array([0.3, -0.5]), np.array([0.1, 0.2])
r.rne(q, qd, np.zeros(2))           # OK
r.copy().rne(q, qd, np.zeros(2))    # ValueError: Link(...) is not in list
r.coriolis(q, qd)                   # ValueError: Link(...) is not in list

Notes

• Dynamics functions on a model with no dynamic parameters: silent zeros and several different errors; add a clear check #722 lists coriolis -> ValueError under "models with no dynamic parameters". That attribution is wrong: this is the error, and it happens on models that do have dynamics.
• Same family as payload(): Robot-vs-Link ownership doesn't survive copy(), and has no answer for multiple end-effectors #705: state that does not survive copy().
• Workaround: C(q, qd) qd = rne(q, qd, 0) - gravload(q) + friction(qd). Checked against coriolis on the DH Puma 560 (max difference 4e-15), and against a Christoffel-symbol calculation from the mass matrix on the URDF robot.

Fix: have Robot.__deepcopy__ remap each link's parent (and children) onto the copied links, and add a regression test that copy() then rne equals the original.

Found while answering Adineh94's question in discussion #158.