From Numerical Interpolation to Constructing Intelligent Behaviors

Jianwei Zhang, Alois Knoll

Abstract: In this paper we propose a general framework for describing and constructing sensor-based behaviours. We point out that a complex high-level task can be realised by a set of modular, cooperating behaviours. Each of these behaviours can be decomposed into local control actions which can be interpreted using linguistic “IF-THEN” rules. Several sample applications in real-world robotic systems are presented: a mobile gripper system equipped with proximity sensors and a two arm system with force/torque sensors. For controllers without sensor-action models, this framework can be universally applied after properly selecting the system inputs. Furthermore, “common sense” knowledge can be integrated and the control parameters can be rapidly adapted through incremental learning.