Combining RayPerception and Camera Sensor in the same agent

I'm working on a project to simulate pedestrian behavior in space (maze, obstacle fields, building floor plan, cities etc...). I train agent with Ray Perception and another with Camera Sensor, and a third one combining both. The problem the camera sensor that it's too computationally heavy, but I'm trying to understand if giving the agent image of the space it will learn better to understand what the space is made from (how an opening in the wall looks like so it understand to go through them and not get stuck in the same room).

Is this an overkill to use both? Are there any significant different in the agent's behavior when using on or the other?

 

Also, is training agents on multiple environments (different layouts for instance) at the same time have different effect from training them on one environment, them moving on to the second and so on?