Automatic Safe Behaviour Tree Synthesis for Autonomous Agents

This paper proposes a systematic approach for automatically generating a safe behaviour tree (BT) coordinator to guide an autonomous vehicle to satisfy the goal of a mission without violating the mission safety requirements. The autonomous agent is assumed to have different capabilities to execute multiple actions and operate in a dynamic environment with moving obstacles. For this purpose, we develop a hierarchical and modular synthesis technique that generates safe and reactive Behaviour Trees (BTs) composed of verified actions. At a high level, the synthesized BTs generate a sequence of actions to meet the goals of the mission. Then, the safety issues for the generated BTs are addressed both at the action level by using dynamic differential logic (DL) and globally at the task (BTs) level by forcing a sequence of actions to respect the safety requirement of the successor's action. Adopting the compositional property of the DL allows us to infer the safety of a system from validating its components. Therefore, taking advantages of DL's compostitional property and BT's modular reactivity property, the proposed algorithm synthesizes a safe and correct sequence of actions that can meet the goal and safety requirements of a mission. The details of the developed algorithms are illustrated through an example, verifying the effectiveness of the proposed approach.