In jacobian.py, f.joint.axis.repeat(N, 1) result in shape (N,3) and cannot multiply with the (N,3,3) rotation matrix. Instead it should repeat(N,1,1) to get shape (N,1,3) to multiply with (N,3,3).
In frame.py, the expression theta * self.joint.axis is not permitted since the tensors have shapes (N,) and (3,). unsqueeze on theta solves it.
There is a mistake in the parallel Jacobian calculation for prismatic joints.
(Was mentioned in https://github.com/UM-ARM-Lab/pytorch_kinematics/issues/3 but not fixed)
In
jacobian.py,f.joint.axis.repeat(N, 1)result in shape(N,3)and cannot multiply with the(N,3,3)rotation matrix. Instead it shouldrepeat(N,1,1)to get shape(N,1,3)to multiply with(N,3,3).In
frame.py, the expressiontheta * self.joint.axisis not permitted since the tensors have shapes(N,)and(3,).unsqueezeon theta solves it.