Abstract
Humanoid robots have shown success in many domestic applications. State-of-the-art robots can deliver drinks, fold laundry, cook meals, and even automatically plug themselves in for charging. When developing a new robot system capable of executing such tasks, many di erent subsystems must work in concert, such as sensors for perception, drivers for low-level actuation, and some kind of intelligent, task-executive control. In this research, a situational awareness framework is proposed that combines these elements within an information fusion hierarchy to solve the problem of complex task execution. The framework is implemented on a H20 humanoid robot using the Robot Operating System (ROS) for system management and using the Point Cloud Library (PCL) for perception algorithms. The framework represents the overall control of the robot through an intuitive layering of low-level sensor readings to high-level action execution and establishes a proof-of-concept pick-and-place behaviour built within a vision-based Learning from Demonstration (LFD) architecture.
