How lardemua/atom works
Project type and purpose
• ATOM Calibration is a ROS-based multi-package framework for robotic multi-sensor, multi-modal calibration. It comprises libraries, calibration engines, batch execution, evaluation scripts, message definitions, RViz plugins, simulation worlds, and examples.
Architectural overview
• All packages live in a single Catkin/CMake workspace. They build into ROS nodes, libraries, message/service definitions, and plugins.
• Core abstraction: the ROS middleware bus—topics, services, parameters—connects sensor data, calibration engines, evaluation scripts, and visualization tools.
Main components
A. atom_core (Python/C++ library)
• Provides atomic transformation math, dataset I/O, utilities (cache, drawing, geometry, optimization helpers).
• Used by atom_calibration, atom_batch_execution, atom_evaluation.
B. atom_msgs (ROS msgs & services)
• Defines custom messages (DepthLabels, Lidar3DLabels, etc.) and services for dataset retrieval, interactive markers, saving/loading collections.
• Acts as the contract between nodes and scripts.
C. atom_calibration (Calibration engine & CLI)
• Contains ROS nodes/scripts that implement calibration routines (objective functions, TF estimation, initial guesses, transform correction).
• Subdirectories: collect (data collectors), calibration (optimizer), dataset_playback (replay), initial_estimate, odometry_inspector, transform_corrector.
• Inputs: raw sensor data (rosbags), pattern definitions. Outputs: optimized transformations, YAML/JSON result files.
D. atom_batch_execution
• Automates running multiple calibration experiments (templates, data files, batch scripts).
• Produces CSV and JSON result sets.
E. atom_evaluation
• Scripts for quantitative evaluation of calibration results (RGB-to-RGB, LiDAR-to-Depth, etc.).
• Reads JSON/YAML calibration outputs, computes error metrics, exports CSV.
F. atom_rviz_plugins
• RViz plugin (selected_points_publisher) for interactive pattern selection in point clouds/images.
• Loaded into RViz for manual data collection steps.
G. atom_worlds
• Gazebo worlds, pattern models (charuco, chessboard), and launch files for simulation-based data collection and calibration.
H. atom_examples
• Bringup, calibration, visualization launch files and URDFs for reference robots (rrbot, softbot, mrbot, etc.).
• Provide end-to-end demonstrations.
I. External tools
• ROS core (roscore, rosbag), Gazebo, RViz.
Component interactions and data flow
1. Launch Gazebo world (atom_worlds) via ROS → simulated sensor topics
2. Run dataset recording node/scripts (atom_calibration/collect) → rosbag files
3. Replay dataset with dataset_playback (atom_calibration) → inputs to optimizer
4. Run calibration node (atom_calibration/calibration) → writes optimized TFs (YAML/JSON)
5. Optionally run transform_corrector or odometry_inspector for refinement
6. Run evaluation scripts (atom_evaluation) on results → CSV error reports
Batch mode: atom_batch_execution orchestrates steps 2–5 automatically over multiple datasets.
Architectural patterns and principles
• Modular ROS packages with clear separation of concerns (core library vs. calibration logic vs. batch orchestration vs. evaluation vs. visualization).
• Publish-subscribe / request-reply (services) interactions over ROS.
• Template-based launch/parameter configuration using Jinja2 templates (.j2 files).
• Data flow: raw data → processing → optimization → evaluation → visualization.