Imitation-Based Task Programming on a Low-Cost Humanoid Robot

Humanoid robots are complex service platforms with anthropomorphic features, specifically designed for close interaction with humans. Conventional programming strategies are hardly applicable to humanoids due to the high number of degrees of freedom that must be coordinated concurrently. Therefore, exploiting humanoids’ potential in service tasks remains an elusive goal. One of the most promising techniques for dealing with humanoid robots is programming by demonstration, which allows even unexperienced users to easily interact with the robot based on the teaching by showing or imitation paradigm. In particular, the ability to imitate human gestures and follow task-relevant paths are essential skills for legged humanoid robots, as they provide the fundamental techniques for physical human-robot interaction. This chapter investigates the potential of imitation in programming humanoid motor skills. As target platform, we have adapted a Robosapien V2 (RSV2), a low-cost small humanoid available in the toy market. The chapter focuses on the teaching of basic, humanoid-relevant skills such as body postures and walking paths. We have explored and combined multiple sensing sources to capture human motion for imitation purposes, namely a dataglove, an electromagnetic motion tracker, and a monocular vision system for landmark recognition. The imitation approach illustrated in this chapter is rather general, even though its implementation is constrained by limitations of RSV2 and by sensor inaccuracies. In particular, the chapter reports successful experiments on gesture imitation, including arms motion as well as upper body and head movements. The gesture repertoire learned by the robot can serve both as a body language for understanding human requests in human-robot interaction and as a set of primitives which can be combined for programming more complex tasks. We believe that a deep assessment of a low-cost humanoid robot is extremely important for the robotic research community since the technological requirements and the costs to develop more advanced humanoid robots still prevent them to become broadly available. Currently, most high-end humanoids are developed as prototypes platforms under the supervision of important private companies. Therefore, low-cost humanoid platforms such as RSV2 provide an exciting and affordable opportunity for research in humanoid integration in service tasks.