README.md



Approximated Abstract Perception Experiments

GEM Cart Simulator
Gazebo Simulation for Lane Detection using Lanenet with GEM cart
https://github.com/GEM-Illinois/POLARIS_GEM_e2/tree/main/polaris_gem_experiments/gem_scenario_runner
Our wrapper for Lanenet as a Python package installable with pip
https://github.com/GEM-Illinois/lanenet-lane-detection

TerraSentia Simulator
Gazebo Simulation scripts for Corn Row Detection using ResNet with a simplified TerraSentia robot
https://gitlab.engr.illinois.edu/aap/terrasentia_simplified/
Repository containing required Gazebo and ROS packages for the simplified TerraSentia (also with plenty of packages unrelated to simulation)
https://bitbucket.org/hc825b/terrasentia-gazebo

Dataset
The datasets are organized in one folder for each simulator.
For each simulator, we collect data from two types of experiment settings:

Teleporting vehicle to specified groundtruth states and collect perceived
output from vision/nnet pipeline.
Each experiment result is stored in a pickle file with the name pattern
collect_images_YYYY-MM-DD-HH-MM-SS.*.pickle.
Check inspect_collect_images_pickle.ipynb for the data structure stored
in the pickcle file.
Simulation traces of the end-to-end system integrating sensor, perception,
controller, and vehicle dynamics from a specific initial states.
Each experiment result is stored in a folder with the name pattern
sim_traces_YYYY-MM-DD-HH-MM-SS/.
Each folder contains (1) the yaml file specifying the distribution of the initial states with manually annotated simulation scenarios and (2) pickle files recording the traces of groundtruth states and perceived outputs.


Scripts
Scripts to generate Fig 7 in the paper

python3 visualize_tracking.py