1 paper
Alvaro Cauderan, Gauthier Boeshertz, Florian Schwarb +1
We present an algorithm that learns to imitate expert behavior and can transfer to previously unseen domains without retraining. Such an algorithm is extremely relevant in real-wor…