High-Level Multi-Robot Trajectory Planning And
Spurious Behavior Detection

Leveraging our trajectory planner, we also generated a dataset with multiple high-level behavior missions in different environments and proposed a robust embedding technique to extract the underlying semantic information of the high-level behaviors in the trajectories to detect spurious behaviors in this missions. Some high-level behaviors modeled and successfully detected by our transformer based pipeline are: Sequential behaviors, Mutual exclusion paths, Conditional access, Simultaneity  or re-currency...  Low-level behaviors like forbidden zone breaching or inefficient trajectories are almost perfectly detected. 

The plans were also tested in a low-level simulator using both holonomic and differential drive robots, and video demonstrations were generated.