Robotic therapy refers now to a diverse set of technologies and algorithms that can match or improve the clinical benefits achievable with conventional rehabilitation therapies after stroke and other neurologic injuries. However, the principles by which robotic therapy devices can be optimized are still not well understood. Here, I will first briefly overview the evolution of the technology and science of robot-assisted rehabilitation, including the range of control algorithms used. Then, I will describe recent experimental evidence that suggests three neuro-computational mechanisms that determine the effectiveness of robotic therapy: human slacking, Hebbian learning via proprioceptive stimulation, and mechanical modulation of reward. I will conclude by describing recent attempts to enhance the effectiveness of robotic therapy by combining it with neuro-regeneration, and by making it more accessible via “consumer stroke technology.”