Social Mini-Games Library

Rohan Chandra¹, Shubham Singh², Wenhao Luo³, Katia Sycara⁴

¹ University of Virginia, ² University of Texas at Austin, ³ University of Illinois Chicago, ⁴ Carnegie Mellon University

📄 A Survey of Multi-Robot Navigation in Social Mini-Games

---

Quick Start

1. Installation

Clone the repository:

git clone https://github.com/CRAL-UVA/SMGLib.git
cd SMGLib

Install dependencies:

pip install -r requirements.txt

2. Run Your First Simulation

python3 run_simulation.py

This launches an interactive menu where you can:

• Choose a navigation method (Social-ORCA, Social-IMPC-DR, or Social-CADRL)
• Select an environment (doorway, hallway, intersection)
• Configure agents and parameters
• View results and animations

3. Example Results

After running a simulation, you'll see results like:

SOCIAL-IMPC-DR RESULTS
Environment: doorway  Success Rate: 100.0% (2/2)  Makespan: 15.00s  Flow Rate: 0.1667

Agent     TTG  MR     Avg ΔV  Path Dev  Hausdorff
Robot 0   45   1.000  1.576   28.488    0.145   
Robot 1   47   1.044  1.576   28.493    0.146  

---

Walkthrough: Running a Complete Simulation

Step 1: Launch the Simulator

python3 run_simulation.py

Step 2: Select Method and Environment

Welcome to the Multi-Agent Navigation Simulator
=============================================

Available Methods:
1. Social-ORCA
2. Social-IMPC-DR  
3. Social-CADRL

Enter method number (1-3): 2

Available environments:
1. doorway
2. hallway
3. intersection

Enter environment type (1-3): 1

Step 3: Choose Output Format

Output format options:
1. Clean (minimal text output)
2. Verbose (detailed output with explanations)

Enter output format (1-2): 1

Step 4: Configure Simulation (for ORCA)

For Social-ORCA, you'll configure:

• Number of robots (1-4)
• Start and goal positions for each robot
• Environment-specific constraints

Step 5: View Results

The simulation generates:

• Performance metrics displayed in terminal
• Animation files in logs/[method]/animations/

Step 6: Examine Outputs

Terminal Results:

SOCIAL-IMPC-DR RESULTS
Environment: doorway  Success Rate: 100.0% (2/2)  Makespan: 15.00s  Flow Rate: 0.1667

Agent     TTG  MR     Avg ΔV  Path Dev  Hausdorff
Robot 0   45   1.000  1.576   28.488    0.145   
Robot 1   47   1.044  1.576   28.493    0.146  

---

Performance Metrics Explained

Primary Metrics

Metric	Description	Interpretation
Success Rate	Percentage of agents reaching goals	Higher = better deadlock avoidance
Makespan	Time for all agents to complete	Lower = more efficient
Flow Rate	Agents per unit time through bottleneck	Higher = better throughput

Per-Agent Metrics

Metric	Description	Formula
TTG	Time To Goal (steps)	Steps until goal reached
MR	Makespan Ratio	TTG_agent / TTG_fastest
Avg ΔV	Average velocity change	Σ|v(t+1) - v(t)|
Path Dev	Path deviation from nominal	L2 norm of trajectory difference
Hausdorff	Maximum deviation distance	max distance between trajectories

Please refer to the report for further details on the metrics and algorithm details.

Available Methods

1. ORCA: Optimal Reciprocal Collision Avoidance
2. IMPC-DR - Deadlock Resolution and Recursive Feasibility in MPC-based Multi-robot Trajectory Generation
3. CADRL - Collision Avoidance with Deep Reinforcement Learning

Supported Environments

1. Doorway Scenario.
2. Hallway Scenario
3. Intersection Scenario.

---

Please cite the following work:

@article{chandra2025multi,
  title   = {Multi-robot navigation in social mini-games: Definitions, taxonomy, and algorithms},
  author  = {Chandra, Rohan and Singh, Shubham and Luo, Wenhao and Sycara, Katia},
  journal = {arXiv preprint arXiv:2508.13459},
  year    = {2025}
}