Interactive Task Learning with Discrete and Continuous Features.

摘要: Learning tasks from demonstration is key to the flexibility of robots and their accessibility non-programmers. We present a task learning framework that combines strengths discrete continuous representations. The robot learns set criteria expectations represent goal demonstrated task. consists performing actions fulfill on objects meet criteria. propose modeling with Gaussian distributions. To deal simultaneous multiple tasks, we assume can be multi-modal model them as mixtures Gaussians. an implementation this