The sport log rolling, has interesting dynamic properties, as competitors are coupled through a free floating log. In order to understand the dynamics, humans are modeled by simple mechanical robots, and different analysis are conducted. Through frequency analysis, it can be shown that the coupling between these robots is limited. Furthermore, the control action that affects a competitors most, is also the action that has largest affects on a robot itself. It is shown through calculation of control invariant sets, that there is no safe set in the state space, i.e. a competitor can always win, if applying the right input to robot-log system. The robots are stabilized using a linear full-state feedback controller, found from Linear Quadratic Regulator. Model Predictive Control is introduced, used in battle between the robots. A control algorithm is found, for which one robot can win over another. To test the control algorithm found for the Human-like robots, an experimental setup have been designed. Simulations show good result for the setup, but have yet to be tested in reality.