Number of humans in 'train' and 'test'

[Yuanzizizi]

Excuse me.

I'm confused about the human_num in crowd_sim.py. Why should we set human_num = 1 when self.robot.policy.multiagent_training = 1?

`

np.random.seed(counter_offset[phase] + self.case_counter[phase])
if phase in ['train', 'val']:        
    human_num = self.human_num if self.robot.policy.multiagent_training else 1
    self.generate_random_human_position(human_num=human_num, rule=self.train_val_sim)
else:
    self.generate_random_human_position(human_num=self.human_num, rule=self.test_sim)
# case_counter is always between 0 and case_size[phase]
self.case_counter[phase] = (self.case_counter[phase] + 1) % self.case_size[phase]

` what I want to know is In training stage, do we use 1 human agent or 5 human agents (env.config [sim] human_num=5)?

I really need your help, thanks~

[ChanganVR]

Multiagent-training differentiates CADRL from our policy. The former takes one human during training and ours takes 5 during training. At test time, both are tested on 5-human scenario.

[Yuanzizizi]

Multiagent-training differentiates CADRL from our policy. The former takes one human during training and ours takes 5 during training. At test time, both are tested on 5-human scenario.

Oh, I see❤. Thanks. But what makes the difference? I suppose that, In theory, cadrl could also have five people in training stage?

[ChanganVR]

yeah possibly. I was just reproducing the training detailed in the paper