ReBeT: Architecture-based Self-adaptation of Robotic Systems through Behavior Trees

Robotics software needs to be self-adaptive. Self-adaptation in robotics can, among others, take the form of changing a robot’s task plan or its software architecture at runtime. The latter has shown to be effective in satisfying quality requirements such as minimizing energy consumption and operating safely. However, most self-adaptive robotic systems perform architecture-based self-adaptation to meet the functional goal of completing an assigned mission. Additionally, the mechanisms to accomplish architectural adaptations are mostly adhoc and not oriented towards reuse. We in turn investigate how quality requirements and architecture-based self-adaptation can be facilitated in robotics software while integrating into existing practices to promote practitioners’ adoption and reuse. To this end, we design and implement an extension to the Behavior Trees (BTs) task plan formalism which introduces an explicit consideration of quality requirements. Additionally, we implement a general architectural adaptation layer for ROS2 systems and an extension to BTs which showcases its utilization. Finally, we perform quantitative experiments to evaluate the effectiveness of our approach in satisfying quality requirements via architectural adaptation on a mobile terrestrial robot. We find our approach to be an effective means to address a variety of self-adaptation scenarios within the mission of the system.