All machine learning relevant data from the simulation can easily be configured with in what format the data should be saved, where to saved it and how often.

(No images available yet)

• Pathfinding with collision avoidance
• Automated grasping of objects

In the simulation environment it is possible to have multiple cameras, each with a different configurations for ease of collecting data for different purposes.

Regular Red-Green-Blue colorised images globally light with shadows and reflections

Each class of object in the scene can be given a unique distinguishable color, this eases identifying the different classes of objects from each other in the scene and their location.

Similar to the segmentation camera configuration but instead now every instance of any class has a different color. (Note: missing "cleverer" coloring feature cause unwanted similarity of neighbour colors might occur... NP-Hard graph coloring problem)

(No animation available yet)

Darker shades are further away from camera while conversely the brighter are closer to the camera, useful for figure relative distances in the environment

For more realistic depth images, this camera will provide information on overexposure and shadows due displacement of light source.

• Guassian Noise sampling in data collection.
• Be able to adjust a noise scale parameter, to inject noise into the data collected in multiple stages(camera images, positions of objects and so on).
• Included the new instance segmentation camera from the droid package.

• agent
• neo
• droid