A Behavior Generation Framework for Robots to Learn from Demonstrations

This paper proposes a framework of generating behavior sequences for robots, especially for humanoid robots, to perform complex tasks. This framework provides a method for the robot to generalize common features of demonstrated behaviors, to store the learned behaviors in the memory system, to construct a behavior graph to describe relationships among learned behaviors, to find and assemble a behavior sequence, and to generate similar motion trajectories of basic behaviors when it is placed in a similar but slightly different task-relevant situation. Additionally, we successfully use behavior graph to describe the dynamic relationship among behaviors and we successfully apply shortest path searching methods in behavior sequence generation, which provides a novel solution to associate knowledge representation with behavior generation. Simulation and experiments are carried on a humanoid robot to validate our proposed framework.