Generic approach for capturing datasets

[fred-labs]

Currently, the only supported method for capturing data with scenario execution is using ROS2 (using the action record_bag).

We want to support at least one other simulation, e.g. pybullet and provide a (generic) approach to capture data.

Example:

import osc.helpers
import osc.pybullet
import osc.data_writer # <-- new library

scenario example_pybullet:
    timeout(60s)
    sim: simulation_pybullet
    actor1: actor_pybullet
    writer: kitti_writer("./my_dataset", "odometry1") # <-- new actor
    do serial:
        sim.initialize("plane.urdf")
        actor1.spawn("husky/husky.urdf", pose_3d(position: position_3d(z: 0.5)))
        parallel:
            sim.run()
            writer.capture() # <-- new action
            actor1.set_joint_motor_control(5mps, 1)
            serial:
                actor1.distance_traveled(2m)
                emit end

In this example only the odometry might be captured (as no image provider is instantiated).

Different data annotation formats might be supported, e.g.:

• kitti
• csv/jpg
• ...

Stretch Goal

Create example on how to use captured dataset to train an AI.