Humanoid robots have been developed with the intention of aiding in environments designed for humans. As such, the control of humanoid morphology and effectiveness of human robot interaction form the two principal research issues for deploying these robots in the real world. In this thesis work, the issue of humanoid control is coupled with human robot interaction under the framework of scaled autonomy, where the human and robot exchange levels of control depending on the environment and task at hand. This scaled autonomy is approached with control algorithms for reactive stabilization of human commands and planned trajectories that encode semantically meaningful motion preferences in a sequential convex optimization framework. The control and planning algorithms have been extensively tested in the field for robustness and system verification. The RoboCup competition provides a benchmark competition for autonomous agents that are trained with a human supervisor. The kid-sized and adult-sized humanoid robots coordinate over a noisy network in a known environment with adversarial opponents, and the software and routines in this work allowed for five consecutive championships. Furthermore, the motion planning and user interfaces developed in the work have been tested in the noisy network of the DARPA Robotics Challenge (DRC) Trials and Finals in an unknown environment. Overall, the ability to extend simplified locomotion models to aid in semi-autonomous manipulation allows untrained humans to operate complex, high dimensional robots. This represents another step in the path to deploying humanoids in the real world, based on the low dimensional motion abstractions and proven performance in real world tasks like RoboCup and the DRC.