MRTA-Benchmark dataset: 250K Optimal Multi-Robot Task Allocation Instances with Heterogeneous Robots, Precedence Constraints & Dynamic Coalitions

Author: Jakob Bichler
Contact: j.d.bichler@gmail.com
Institution: TU Delft – MSc Robotics, Autonomous Multi-Robot Lab
License: CC BY 4.0

---

📦 General Introduction

This dataset contains 250,000 optimally solved instances of multi-robot task assignment problems with heterogeneous robots and dynamic coalition formation. It was created as part of the Sadcher scheduling framework, developed during a master thesis at TU Delft.

Each problem instance specifies task requirements, robot capabilities, spatial task layouts, precedence constraints, and travel/execution times. Each corresponding solution file describes the optimal schedule minimizing the makespan, generated using a MILP solver.

---

📁 File Structure and Naming Convention

• problems/: JSON files with input data for each instance.
• solutions/: JSON files with corresponding optimal task schedules.

Filenames are indexed numerically and match across folders, e.g.:

problems/instance_000001.json
solutions/instance_000001.json

---

🧪 Methodological Details

• Instance generation: Randomized task/robot features sampled under constraints to ensure full skill coverage and valid scheduling.
• Solving method: Mixed Integer Linear Programming (MILP) using PuLP and CBC solver.
• Resources: Dataset generated using DelftBlue HPC cluster (48 cores).
• Software:
  • Python 3.9+
  • PuLP (MILP modeling)
  • CBC solver
  • NumPy, Matplotlib (visualization)

---

📊 File Format Description

🔹 Problem Instance (problems/)

JSON object with:

• Q: Robot skill matrix, shape (N x S), binary.
• R: Task requirement matrix, shape ((M+2) x S), binary, includes start/end dummy tasks.
• T_e: Execution times per task, length M+2.
• T_t: Travel time matrix, shape (M+2 x M+2), symmetric.
• task_locations: List of [x, y] coordinates, length M+2.
• precedence_constraints: List of [i, j] pairs enforcing task i before j.

🔹 Solution File (solutions/)

JSON object with:

• makespan: Total schedule duration.
• n_tasks, n_robots: Problem size.
• robot_schedules: Dictionary mapping robot IDs to:
  • task: Index of the assigned task.
  • start_time, end_time: Scheduled time window.

---

🧮 Units and Conventions

• Distances: Euclidean; unitless (assumed meters if needed).
• Times: Arbitrary unit (can be scaled); consistent across execution and travel times.
• Skills: Binary vectors; 1 indicates possession or requirement.
• Missing data: Not applicable—instances are fully specified and validated.

---

🔐 License

This dataset is released under the CC BY 4.0 License. You are free to use, adapt, and distribute it with proper attribution.

---

For questions or suggestions, please contact: j.d.bichler@gmail.com