The most difficult issue in applying either ASSF, EPB/PDO, or any two dimensional qualitative representation to a real robot reasoning system is the extension of the methods developed to three dimensions. The issues involved are too complex to discuss in any depth, but some ideas of initial directions to investigate include the following:

- Both generalised cones and boundary representation solid modelling techniques are based on two dimensional descriptions at a certain level. A boundary representation must include 2D faces as part of the boundary, and the generalised cone technique requires a method for describing two dimensional cross-sections.
- The majority of actions in a mechanical workspace involve simultaneous motion in no more than two dimensions. People also appear to have difficulty reasoning about motion in three independent dimensions, so limiting a robot reasoning system to two dimensions for actual motion planning may be an acceptable restriction if human-like performance is the goal. These two dimensions might be in either a cartesian or polar co-ordinate system, since simultaneous turning and linear motion in the case of screw operation is not a difficult reasoning problem for people.
- In two dimensions, there are only three types of possible contacts between object boundaries: junction-junction, segment-segment, and junction-segment. In three dimensions, there are six types of possible contact: plane-plane, plane-edge, plane-vertex, edge-edge, edge-vertex and vertex-vertex. This requires that planes be explicitly represented, and also that methods of expressing relative orientation and contact be more sophisticated. This would require a completely different approach to defining proximity (which must reflect potential contact).
- Wong and Fu describe a method for planning motion in three dimensions, operating in the space of a two-dimensional orthogonal projection [WF86].