Hi,

Thank you for the excellent work.

I'm running the default configuration with g1_ground and found that _reward_feet_distance() is a little weird, as shown below:

    def _reward_feet_distance(self):
        left_foot_pos = self.rigid_body_states[:, self.left_foot_indices, :3].clone()
        right_foot_pos = self.rigid_body_states[:, self.right_foot_indices,  #:3].clone()
        feet_distances = torch.norm(left_foot_pos - right_foot_pos, dim=-1)
        reward = tolerance(feet_distances, [0, 0.4], 0.38, 0.05)
        return (feet_distances > 0.9).squeeze(1)

The function seems to encourage the feet to maintain a certain distance, but it returns (feet_distances > 0.9).squeeze(1), which is not related to the previous computation. Should it return reward instead?

I have no idea whether I'm misunderstanding the function or it's a bug? If this is intentional, could you explain the design?

Thanks for your time, and for the great work on this project.