Agile robotic planning with Gwendolen

The future of robotics foresees autonomous behavior that can complete tasks intelligently, with a focus on adaptability, flexibility, and versatility. In such systems, it is critical for robots to quickly and safely perform an operation. However, such aptitude is not limited to the speed of solving tasks, but also requires other qualities such as adeptly detecting and recovering from task irregularities, overcoming unforeseen task barriers by replanning to achieve stated goals, and adroitly adapting to dynamic environments such as changing light illumination, noisy sensors, or unexpected conditions. These intelligent characteristics define robot agility (not to be confused with robot agility akin to dexterity), and refer to approaches that allow robotic systems to be flexible and capable of re-tasking in the face of a changing and often unpredictable environment. Because robot task agility requires sophisticated dynamic and continuous planning and replanning, the Gwendolen intelligent agent programming language is studied as a high-level robot planner. In this report, we develop a manufacturing kitting case study to research the operation of Gwendolen planning. The case study uses the combination of Gwendolen, Canonical Robot Command Language (CRCL), Robot Operating System (ROS), and Gazebo software components to simulate and evaluate robot planning. Several Agile Robotics for Industrial Applications Competition (ARIAC) kitting agility challenges are used to evaluate Gwendolen planning under various levels of operational duress. Both the benefits and shortcomings will be reviewed.