Embedded simulator as benchmark data input

[hidmic]

Feature description

At present, LAMBKIN relies on pre-recorded bagfiles as benchmark data input. It is up to the user to collect them, using actual hardware or in simulation. For the latter, embedding simulated data generation would significantly improve LAMBKIN's usability and we could leverage any of the simulators that have emerged from the embodied AI community.

Implementation considerations

iGibson seems like a great match. It offers many scene variations out of the box, which it can further increase by domain randomization, it offers simple APIs for scene traversal, it offers plenty sensor modalities, it partiallly integrates with ROS, it's all Python.