I've been trying to calculate road for a steeply banked oval track and the algorithm falls apart. Because the outer edge typically has an abrupt change, the algorithm generates waves where the aerial LiDAR scan has gaps. Increasing panel length helps, but you still end up with odd panel vertices a meter or more below the scan even with the panel length being 8x lengthier than the scan intervals. Increasing the sample radius does worse things. Your entire banking won't match the scan data, so 15 degree banking ends up being about 9 degrees. In the end, I can only suggest that you model ovals and save the aerial LiDAR for infield road courses. I suspect if the scan data were more dense by an order of magnitude or two, then the results would be much better.