We present a new approach, called controlled linear perturbation (CLP), to the robustness problem in computational geometry and demonstrate it on Minkowski sums of polyhedra. The robustness problem is how to implement real RAM algorithms accurately and efficiently using computer arithmetic. Large errors can occur when predicates are assigned inconsistent truth values because the computation assigns incorrect signs to the associated polynomials. CLP enforces consistency by performing a small input perturbation, which it computes using differential calculus. CLP enables us to compute Minkowski sums via convex convolution, whereas prior work uses convex decomposition, which has far greater complexity. Our program is fast and accurate even on inputs with many degeneracies.