Why the policy is implemented using a multilayer perceptron (MLP) composed of one layers

[zxz330]

class G1RoughCfgPPO( LeggedRobotCfgPPO ): class policy: init_noise_std = 0.8 actor_hidden_dims = [32] critic_hidden_dims = [32] activation = 'elu' # can be elu, relu, selu, crelu, lrelu, tanh, sigmoid # only for 'ActorCriticRecurrent': rnn_type = 'lstm' rnn_hidden_size = 64 rnn_num_layers = 1

[forresti]

I have the same question.

[wangDx1221]

maybe just provide a sample example.