[GeeTwo]

Is the intention for this to be a "blind" robot that follows a single path to the end point that we're trying to optimize, or a "planner" that virtually follows many routes and then selects the "shortest" one?
If a planner, does the planning algorithm "know" a priori where the obstacles are (that is, it can directly access the table of obstacle positions, rather than having to virtually bump in to them)?

How large does this have to scale? That is, are you looking at "Big O" performance of robots with 100 degrees of freedom and 10,000 obstacles, or are we more likely to be solving problems with 2 to 7 DoF and a dozen obstacles? Optimizing for these cases is very different.

Are we considering a cost function, or just trying to find some route? If so, is the cost only a function of "distance" traveled, number of "course changes", or are speed/acceleration taken into account? Is cost a function of "location"? Is the cost isotropic (are the dimensions scaled so that moving 1 unit in each dimension has the same constant cost)? If you move in multiple dimensions at the same time, how do you calculate the cost? (e.g.: sum of distances along each axis, pythagorean distance, or something else)?

If two obstacles in n-space share a "face segment" of n-1 dimensions, can the robot move along that interface? For example, in 2-space, if one obstacle is (0-1, 1-3), and another is (1-2,0-2), can we move along the path (1,1) to (1,2)? (My guess is not, but I wanted to check.)