Robot Navigation in Crowd Based on Dual Social Attention Deep Reinforcement Learning

Algorithm 1 Value network training.

(1)	Initialize experience replay memory with demonstration
(2)	Initialize value network with memory
(3)	Initialize target value network
(4)	fordo
(5)	Initialize joint state
(6)	repeat
(7)
(8)
(9)
(10)	Enrich experience
(11)	Optimize value network with experience
(12)	Update value network by gradient descent
(13)	until terminal state or
(14)	Update target network
(15)	end for
(16)	return