Using this catkin workspace, I run it as follows: 0. Build this catkin workspace with catkin build and source with source devel/setup.bash. Source in all terminals.

1. Terminal 1: run roscore
2. Terminal 2: run rosparam set use_sim_time true
3. Terminal 2: Run a rosbag from lester in the background. I use this loop one: s3://arl-aimm-data/warthog/2023-05-05/gq_cmu_forest_loop_data_collect_04_2023-05-05-13-40-11.bag with the following command: rosbag play --clock gq_cmu_forest_loop_data_collect_04_2023-05-05-13-40-11.bag
4. Terminal 3: run roslaunch learned_cost_map cmu_sara_stack_lester.launch
5. Terminal 4: run python robot_dataset/scripts/online_hdif.py. Make sure to set the right parameters for where the pre-trained models are stored and where the models will be stored. These should ideally be put into an argparser or a hydra config.
6. Make sure you can visualize training loss with wandb.