Towards Adaptable and Uncertainty-Aware Behavior Trees

Paper i proceeding, 2025

Space robotic missions are taken on a highly uncertain ground, yet require high autonomy. In space, events are unknown and their effects are hard to predict. Mission designers are forced to make decisions despite an inherent lack of information and this results in complex and stiff specifications. Stiffness flags for brittleness. Towards flexibility and modularity, Behavior Trees foster a tractable notation for reactive behavior, attracting the spotlight of robotic mission specifications. However, they lack support for taming uncertainty at runtime. This paper proposes a first step towards the extension of behavior trees with adaptability in order to deal with uncertainty. Our implementation extends the behavior trees constructs with adaptable nodes, i.e., nodes that can be hot-swapped at runtime. Our framework relies on quasi-natural language requirements modeling in FRETISH notation, with transformations to uncertainty-aware behavior trees and deployment to space robotics scenarios in the context of Space ROS. We showcase the use of our framework within the simulation of a NASA mission on Mars.