Computational geometry algorithms are hard to implement robustly and efficiently. One problem is that the control logic is expressed in terms of the signs of predicate polynomials. The polynomials must be evaluated exactly, since any numerical error can cause an incorrect sign and hence a large output error. Degenerate (zero) values must be detected and they often entail special case logic. Bit complexity must be controlled in iterated computations by rounding, typically to double float, while preserving combinatorial properties. Curved (nonlinear) geometry must be manipulated despite its high algebraic degree. Milenkovic and I have solved some of these problems and are working on the others.