Learning swarm behaviour from a flock of homing pigeons using inverse optimal control
This work provides a method to infer individual decision-making in animal swarms from trajectory data, relevant for biologists and robotics researchers studying collective behavior.
The authors analyzed GPS data from homing pigeons to model flocking behavior as a swarm optimal trajectory tracking control problem, where followers track a leader ahead. They proposed a minimum principle-based method to learn unknown cost function weights from flight data, enabling inverse optimal control for each follower.
In this work, Global Position System (GPS) data from a flock of homing pigeons are analysed. The flocking behaviour of the considered homing pigeons is formulated as a swarm optimal trajectory tracking control problem. The swarm problem in this work is modeled with the idea that one or two pigeons at the forefront lead the flock. Each follower pigeon is assumed to follow a leader pigeon immediately ahead of themselves, instead of directly following the leaders at the forefront of the flock. The trajectory of each follower pigeon is assumed to be a solution of an optimal trajectory tracking control problem. An optimal control problem framework is created for each follower pigeon. An important aspect of an optimal control problem is the cost function. A minimum principle based method for multiple flight data is proposed, which can help in learning the unknown weights of the cost function of the optimal trajectory tracking control problem for each follower pigeon, from flight trajectories' information obtained from GPS data.