A multi-input multi-output (MIMO) algorithm based on higher-order Poincar'e spheres is demonstrated for space-division multiplexing (SDM) systems. The MIMO algorithm is modulation format agnostic, robust to frequency offset and does not require training sequences. In this approach, the space-multiplexed signal is decomposed in sets of two tributary signals, with each set represented in a higher-order Poincar'e sphere. For any arbitrary complex modulation format, the samples of two tributaries can be represented in a given higher-order Poincar'e sphere with a symmetry plane. The crosstalk along propagation changes the spatial orientation of this plane and, therefore, it can be compensated by computing and realigning the best fit plane. We show how the transmitted signal can be successfully recovered using this procedure for all possible combinations of tributaries. Moreover, we analyze the convergence speed for the MIMO technique considering several optical-to-noise ratios.