Behavior-Based Visual Servoing on a Hybrid CPU-/FPGA-Platform

Abstract:
Visual servoing for the interaction of mobile manipulators with their environment has many applications throughout research and industry. What these often lack is to achieve both a predictable, robust control for safe robot arm movements and implementation flexibility. Based on the iB2C-paradigm, this work presents an entirely behavior-based architecture distributed between the FPGA and processor parts of a SoPC. This combination allows the exploitation of advantages of both worlds. A behavior network realizing reactive kinematics and target approach is realized on the FPGA. Its hard real-time constraints, deterministic behavior, and quick update rates enable a safe execution. On a higher perception and control level, a second behavioral sub-network implemented in software on top of an OS is executed by the hard processor. This enables the use of open-source software and advanced programming languages as well as middleware, analysis, and supervision software. An interface abstraction tailored for this structure is introduced to satisfy the need for fast communication between the two computing units. The approach is validated on a 5-DOF robot arm.