A predictive control solver for low-precision data representation

We propose a method to efficiently exploit the non- standard number representation of some embedded computer architectures for the solution of constrained LQR problems. The resulting quadratic programming problem is formulated to include auxiliary decision variables as well as the inputs and states. The new formulation introduces smaller roundoff errors in the optimization solver, hence allowing one to trade off the number of bits used for data representation against speed and/or hardware resources. Interestingly, because of the data dependencies of the operations, the algorithm complexity (in terms of computation time and hardware resources) does not increase despite the larger number of decision variables.