In most countries, a project for the development of an urban area has to obey zoning regulations. In France, such zoning regulations are specified in local urban planning schemes (LUPS or PLU in French) defining the right to build at the scale of a parcel. Such rules define, for example, the maximal building height. As the rules are stated in technical documents, they are not easy for non-professionals to comprehend. It is also hard for professionals to assess their impacts. Driven by such issues, we propose to generate 3D building layouts that comply with these rules while optimizing urban indicators (e.g. floor area ratio). A building layout can be seen as a realization of a marked point process (MPP), which is a stochastic model mapping from a probability space to configurations of geometric objects, namely horizontal 3D boxes. Then, the problem of finding an optimized building layout is converted into finding the optimal realization of a MPP of 3D boxes. We solve this optimization problem by trans-dimensional simulated annealing (TDSA), which allows to explore both parameter space and model space in order to find the combination optimizing a given criterion or energy function. A global energy function is defined as the sum of weighted energy terms. Each energy term is able to penalize the building layouts that violate a specific rule or favor the ones according to the optimization task. TDSA generates the optimal building layout by minimizing this global energy using the coupling of a simulated annealing scheme with a Reversible Jump Markov Chain Monte Carlo (RJMCMC) sampler. We studied several common types of the French PLU rules and modeled them into energy terms. A case study is conducted and the results show that our proposed approach is capable of such an optimization task within a short computation time.