Temporal Logic Motion Planning of a UAV-UGV Team in Semantically Unknown Environments

We address the problem of temporal logic motion planning for a UAV-UGV team operating in an uncertain environment where the semantic labels are not known a priori. For example, consider a UGV that is responsible for collecting samples A, B, and C in a specific order to complete a desired task in a space mission; however, the exact locations of samples are initially unknown, and instead a probabilistic initial belief is available. We propose a planning algorithm for the UAV, whose objective is to assist the UGV by gathering as much information as possible about the environment while satisfying a distance constraint relative to the UGV. The UAV also leverages vision-language models (VLMs) for perception to facilitate detection and interpretation of high-level semantic labels in the environment. We show that the UAV-UGV team is able to accomplish the task safely even under an incorrect initial belief via simulations.