Our current simulation setup seems to be far too perfect to be any good for testing. So the idea would be to intentionally ruin some of the setup in a way that adds noise and uncertainty to the system.
Some ideas:
in real life the lidar and camera extrinsics positions can be widely off, so we need to figure out how to add some angular and positional noise to these positions
lidar data distance accuracy should be lowered
linear odometry should be slightly lowered in accuracy, and angular odometry significantly
Our current simulation setup seems to be far too perfect to be any good for testing. So the idea would be to intentionally ruin some of the setup in a way that adds noise and uncertainty to the system.
Some ideas:
Lidar datasheet, showing range and accuracy: