Symbolic representation of what robots are taught in one demonstration
Andrea Maria Zanchettin
Abstract
To facilitate the use of robots in small and medium-sized enterprises (SMEs), they have to be easily and quickly deployed by non-expert users. Programming by Demonstration (PbD) is considered a fast and intuitive approach to handle this requirement. However, one of the major drawbacks of pure PbD is that it may suffer from poor generalisation capabilities, as it is mainly capable of motion-level representations. This work proposes a method to semantically represent a demonstrated skill, so as to identify the elements of the workspace that are relevant for the characterisation of the skill itself, as well as its preconditions and effects. This way, the robot can automatically abstract from the demonstration and memorise the skill in a more general way. An experimental case study consisting in a manipulation task is reported to validate the approach.