Abstract: We present an end-to-end differentiable learning algorithm for multi-agent navigation policies. Compared with prior model-free learning algorithms, our method leads to a significant speedup ...