Replication data for: Data-Driven Anomaly Detection for Graph-based Formation Control in a Robot Swarm

The simulation of the robot formation, the data preprocessing, and the anomaly detection method are described in the corresponding paper. Each row of the dataset contains a single robot action, which represents the robot's change in position during an interval of 2s, and the corresponding context information of the formation.

The label column identifies the robot as a formation leader (10), a normally behaving single-edge follower (11), a normally behaving double-edge follower (12), or one of the following robot with anomalous behavior: formation-changing robot (2), swinging robot (3), noisy robot (4), or brute-force robot (5). The context information includes the previous robot action previous_action_x|y, the distances to the other robots dx|y_r, the distance to the formation center dx|y_center, and the distance to a look-ahead point on the leader trajectory dx|y_trajectory.