How to make humanoid gaze towards specific object?

[albertcity]

I am trying to use Habitat to generate a multimodal dataset (video+language annotations) which may help improve the spatial and temporal perception abilities of VLMs.
I am using the Humanoid to interact with the environment, such as navigation/pick/place. However, I found when pick some object, the humanoid does not look at the object in interest (from the perspective of head_rgb sensor); the object to pick is missed in the view. Besides, during the pick action, the robot’s head RGB sensor clipped through the model, resulting in a view from inside the robot model.

Is there any method or hint that can solve/bypass the above problem?