MRDG

Multi-Robots Dataset Generator (MRDG) is a generator of datasets for multi-robots scenarios. It can generate data for C-SLAM backend algorithms. MRDG is also able to compute errors from computed results.

---

---

Installation

MRDG runs inside Dev Containers — no manual dependency installation is required. The only prerequisites are Docker, the Dev Containers extension for VS Code and X11 server.

Steps

1. Clone the repository

   git clone https://github.com/tschweitzer57/MRDG.git
   cd MRDG

2. Open in VS Code and select the appropriate Dev Container when prompted (or via Reopen in Container):

   Container	Purpose	Docker image
   dataset-gen	Dataset generation	tschweitzer57/gtsam:base-20.04
   error-metrics	Metrics computation & visualization	tschweitzer57/gtsam:base-20.04
   dataset-gen-v2	Dataset generation (to be removed)	tschweitzer57/gtsam:4.3a0-ros2-jrl
   dataset-gen-v3	Dataset generation (under developement)	tschweitzer57/gtsam:4.3a0-ros2jazzy-mirage
   metrics-v2	Metrics computation & visualization (to be removed)	tschweitzer57/gtsam:4.3a0-ros2-jrl
   metrics-v3	Metrics computation & visualization (under developement)	tschweitzer57/gtsam:4.3a0-ros2jazzy-mirage

3. Wait for the container to build. The workspace, helpers, scripts, configuration files, and output folders are automatically bind-mounted inside the container — no extra setup is needed.

---

Usage

MRDG covers two main workflows: dataset generation and metrics computation.

1. Dataset generation

Open the dataset-gen Dev Container.

Generate a dataset from the default configuration

cd scripts
python gen-default.py

This generates a dataset using 2_Configurations/DEFAULT/default.json (4 robots, 250 poses, 30 landmarks) and saves the output to 4_Datasets/.

Generate datasets from a configuration folder

cd scripts
python gen-vld1.py

This iterates over all .json files in 2_Configurations/VLD1/ and generates one dataset per configuration file.

Parse and visualize a generated dataset

cd scripts
python parse-dataset.py

Configuration format

Configuration files are JSON files located in 2_Configurations/. The main parameters are:

{
  "name": "my_dataset",
  "trajectory": {
    "robots": 4,
    "poses": 250,
    "seed": 0
  },
  "landmarks": {
    "number": 30,
    "detection_prob": 0.8
  },
  "intra-loop-closure": { "number": 10, "frequency": 5 },
  "inter-direct-loop-closure": { "pose": true, "range": true },
  "sigmas": {
    "odom": [0.01, 0.01, 0.01],
    "lc_pose": [0.05, 0.05, 0.05]
  },
  "outliers": {
    "false_matching": 0.05,
    "robot_loss": false
  }
}

See 2_Configurations/DEFAULT/complete.json for the full list of available parameters.

---

2. Metrics computation

Open the error-metrics Dev Container. Place solver results (.jrr.cbor files) in 5_Results/Computed/ before running the scripts.

cd scripts
python TEST_4.py       # Compute errors across solver iterations
python BR_1_display.py # Display swarm consensus and landmark errors

Output metrics are saved to 5_Results/Metrics/ and plots to 5_Results/Display/.

Repository structure

• Documentation : Contains MRDG documentation and images
• Generator : Contains datasets generation helpers and scripts for different scenarios
• Configurations : Contains configuration files for generation helpers and scripts
• Metrics : Contains metrics helpers and script to display errors
• Datasets : Contains generated datasets
• Results : Contains computed results (errors and graphs)

---

Future improvements

Todo list