Predicting Actions to Act Predictably: Cooperative Partial Motion Planning with Maximum Entropy Models