Do what i want, not what i did: Imitation of skills by planning sequences of actions

Abstract

Abstract-We propose a learning-from-demonstration approach for grounding actions from expert data and an algorithm for using these actions to perform a task in new environments. Our approach is based on an application of sampling-based motion planning to search through the tree of discrete, high-level actions constructed from a symbolic representation of a task. Recursive sampling-based planning is used to explore the space of possible continuous-space instantiations of these actions. We demonstrate the utility of our approach with a magnetic structure assembly task, showing that the robot can intelligently select a sequence of actions in different parts of the workspace and in the presence of obstacles. This approach can better adapt to new environments by selecting the correct high-level actions for the particular environment while taking human preferences into account.

A cognitive architecture for automatic gardening

“…For a successful integration of these techniques in a robotic platform, it is mandatory to ground the symbolic descriptions of the AI planning methods to let the robot interact with the real world in order to execute a task [12]. This is normally done by integrating methods of different levels of abstractions [10,13,14,15,16], ranging from purely symbolic methods [17], to sensor information processing [18] and acting methods [19]. Several approaches integrate planning and acting for the robotic execution of human like tasks.…”

“…This is done by exploiting semantic similarities between tasks parameters. Learning from demonstration has also been used in the architecture presented in [16]. In this case, a human provides example executions of symbolic actions that are used to update low-level probabilistic models parametrized for each specific robotic platform, which permits evaluating the feasibility of grounding symbolic actions.…”

Evaluating Methods for End-User Creation of Robot Task Plans

“…These developments have led to a growing interest in making it easy for domain experts to transfer knowledge to collaborative robots, either through a user interface [2], [3], [4], [5], natural language [6], or learning from demonstration [7], [8], [9], [10]. To take full advantage of these systems, the human user must have an accurate mental model of a robot's capabilities [11].…”

“…Recent approaches for end-user instruction of collaborative robots include the development of new user interfaces [2], [3], [4], learning from demonstration [7], [10], or systems that make use of natural language together with ontologies and large knowledge bases to follow high-level instructions, such as Tell Me Dave [6] or RoboSherlock [12].…”

“…Dianov et al [9] take a hybrid approach, using task graph learning to infer task structure from demonstrations and a detailed ontology. Other recent work explored combining learned actions with sampling-based motion planning and a high-level task specification [10].…”