Behavior Trees for Robotics Software Engineering

Doktorsavhandling, 2026

Context. Robotic systems are used more widely in homes and industry, but their often ad hoc development methods reduce the reliability needed for ongoing growth. Recently, more robotics experts have adopted software engineering best practices to improve system reliability. Building on previous work, we aim to further support the development of reliable robotics missions.
Objective. The objective of this thesis is to build systematic knowledge from empirical insights into the current state of practice in specifying robotics missions, and to develop solutions that support improved software engineering practices for the development of robotic missions based on our accumulated knowledge.
Method. This thesis employs a mixed-methods approach, combining empirical and constructive research methodologies. The empirical phase involved knowledge-seeking studies that employed comprehensive analyses of behavior modeling languages and their domain-specific languages, mined open-source projects, and conducted action research and survey studies with practitioners. The constructive research phase involved solution-seeking studies employing the accumulated knowledge to design, implement, and evaluate two solutions in collaboration with practitioners to problems identified in the previous phase. Our research was grounded in the context of robotics software engineering and, when possible, conducted in collaboration with both academic and industrial practitioners.
Result. The empirical contributions provided a comprehensive understanding of behavior trees as a behavior modeling language for robotics missions. This included comparisons with UML behavior modeling languages and state machines, as realized in robotics, regarding modeling concepts and language design, their adoption in open-source robotic projects, and practitioner practices and experiences. Based on these insights, two solutions were developed and evaluated with practitioners.
The first introduces a meta-model extension and domain-specific language for behavior trees, integrated into a popular library, to specify quality concerns. The second proposes an approach for supporting industrial risk assessment by integrating behavior trees, enabling early identification of risk concerns and ensuring that their outcomes are adequately considered in the software implementation of robotics missions.
Conclusion. This thesis advances empirical understanding of behavior modeling languages for specifying robotics missions, focusing on behavior trees, an emerging model for behavior coordination and execution in robotics. The built knowledge about behavior trees characterizes their support, usage, and practitioner perspectives. The developed solutions show behavior trees' potential beyond coordination and execution, using them to capture quality concerns and support risk assessment, thereby opening new directions for their role within the broader robotic software engineering lifecycle.