Admissible configurations

Assume that $ {\cal W}$, $ {\cal O}$, and $ {\cal A}$ from Formulation 4.1 are used. For manipulation planning, $ {\cal A}$ is called the manipulator, and let $ {\cal C}^a$ refer to the manipulator configuration space. Let $ {\cal P}$ denote a part, which is a rigid body modeled in terms of geometric primitives, as described in Section 3.1. It is assumed that $ {\cal P}$ is allowed to undergo rigid-body transformations and will therefore have its own part configuration space, $ {\cal C}^p = SE(2)$ or $ {\cal C}^p = SE(3)$. Let $ {q}^p \in
{\cal C}^p$ denote a part configuration. The transformed part model is denoted as $ {\cal P}(q^p)$.

Figure 7.15: Examples of several important subsets of $ {\cal C}$ for manipulation planning.
\begin{figure}\begin{center}
\begin{tabular}{ccc}
\psfig{file=figs/manip1.eps,wi...
...l C}_{sta}}$ & $q \in {{\cal C}_{tra}}$
\end{tabular}\end{center}
\end{figure}

The combined configuration space, $ {\cal C}$, is defined as the Cartesian product

$\displaystyle {\cal C}= {\cal C}^a \times {\cal C}^p ,$ (7.14)

in which each configuration $ q \in {\cal C}$ is of the form $ q = (q^a,q^p)$. The first step is to remove all configurations that must be avoided. Parts of Figure 7.15 show examples of these sets. Configurations for which the manipulator collides with obstacles are

$\displaystyle {{\cal C}^a_{obs}}= \{ (q^a,q^p) \in {\cal C}\;\vert\; {\cal A}(q^a) \cap {\cal O}\not = \emptyset \} .$ (7.15)

The next logical step is to remove configurations for which the part collides with obstacles. It will make sense to allow the part to ``touch'' the obstacles. For example, this could model a part sitting on a table. Therefore, let

$\displaystyle {{\cal C}^p_{obs}}= \{ (q^a,q^p) \in {\cal C}\;\vert\; \operatorname{int}({\cal P}(q^p)) \cap {\cal O}\not = \emptyset \}$ (7.16)

denote the open set for which the interior of the part intersects $ {\cal O}$. Certainly, if the part penetrates $ {\cal O}$, then the configuration should be avoided.

Consider $ {\cal C}\setminus ({{\cal C}^a_{obs}}\cup {{\cal C}^p_{obs}})$. The configurations that remain ensure that the robot and part do not inappropriately collide with $ {\cal O}$. Next consider the interaction between $ {\cal A}$ and $ {\cal P}$. The manipulator must be allowed to touch the part, but penetration is once again not allowed. Therefore, let

$\displaystyle {{\cal C}^{ap}_{obs}}= \{ (q^a,q^p) \in {\cal C}\;\vert\; {\cal A}(q^a) \cap \operatorname{int}({\cal P}(q^p)) \not = \emptyset \} .$ (7.17)

Removing all of these bad configurations yields

$\displaystyle {{\cal C}_{adm}}= {\cal C}\setminus ({{\cal C}^a_{obs}}\cup {{\cal C}^p_{obs}}\cup {{\cal C}^{ap}_{obs}}) ,$ (7.18)

which is called the set of admissible configurations.

Steven M LaValle 2020-08-14