The basic pseudoinvertion metods of
invers kinematics suffer the problem of non-repeatability of the trajectory
or so-called non-cyclic behaviour. The conditions of repeatability were
putted in mathematical form by Shamir & Yomdin (Shamir, Y. & Yomdin,
Y. (1988) Repeatability of redundant manipulators: Mathematical
solution of the problem. IEEE Transactions on Automatic Control. 33(11),
1004-1009).
We have dq=K*dx, where K
is our inverted J. Collumns of K span a linear N-dimentional
subspace, Range(K), above each non-singular point in joint space.
It would be tangent to our integrable surface, where our joints' trajectories
should be embedded not crossing each other. It is possible if there is
such an integrable surface. Let's derive conditions for existence of integrable
surface.
Suppose that it exists. Let's take
a shift along any direction to the nearest point, being on integrable surface.
We must be sure for our path to be repeatable backward, that each of this
two tangent spaces at two neighbouring points have in common this
shifting vector. So the tangent space at each point of our cyclic trajectory
should contain two neighbouring points of this trajectory, for this
trajectory to be repeatable in each direction. And now let's consize our
cyclic trajectory to very small square of 4 points in joint space,
that includes our "base point".
Let's take two orthogonal shifts,
it will be vectors k indexed i and k indexed j, that start at base
point. So we have a condition for tangent space of each of this 4 points
contains it's two neighbours:
![]()
It is generalized for all pairs of collumns of K due to linearity. This Lie bracket:
