Dynamic Path Planning of Multi-Agent Robotic Systems (MARSs)

This project aims to:

• Develop an adaptive and robust path planning solution for MARSs.
• Bridge the gap between theoretical algorithms and practical industry demands.

By:

Mohammad Ali

In collaboration with:

Adityanshu Abhinav [https://github.com/AadityanshuAbhinav]

Under the Guidance of:

Dr. Anuj Kumar Tiwari

---

Key contributions of this project include:

• Camera Calibration for accurate depth perception and obstacle measurement.
• Simulation using Gazebo to evaluate Bug2-D* Lite before deployment.
• Implementation on TurtleBot3, demonstrating real-world effectiveness.

Potential applications include search and rescue operations, warehouse logistics, and autonomous cleaning tasks. Future enhancements involve handling extreme dynamism and integrating machine learning for advanced object recognition.

---

1. Introduction

1.1 Background and Motivation

Multi-Agent Robotic Systems (MARSs) are revolutionizing industrial automation by enhancing efficiency, productivity, and safety. However, a significant challenge in dynamic path planning arises in unpredictable environments, where obstacles move unexpectedly.

This project aims to:

• Develop an adaptive and robust path planning solution for MARSs.
• Bridge the gap between theoretical algorithms and practical industry demands.

1.2 Research Aims and Objectives

• Design and implement a real-time path planning algorithm incorporating obstacle avoidance and dynamic replanning.
• Evaluate its performance in a simulated warehouse.
• Analyze the algorithm's ability to optimize path efficiency for material transportation.

1.3 Resources & Requirements

Hardware:

• TurtleBot3 Robot with modular design and ROS (Robot Operating System) support.
• LiDAR Sensor (LDS-01) for real-time obstacle detection.
• Intel RealSense Depth Camera D435i for Aruco marker detection and robot localization.
• Aruco Marker Tags for precise robot positioning.

Software:

• ROS Noetic (Ubuntu 20.04 LTS) for robot control and algorithm implementation.
• Gazebo Simulator for testing before real-world deployment.
• Custom ROS Packages implementing Bug2-D* Lite with real-time sensor fusion.

---

2. Literature Review

• Collision Avoidance Algorithms: Utilize velocity obstacles for real-time navigation.
• Multi-Robot Coordination: Centralized vs. decentralized planning approaches.
• Reactive Path Planning: Adaptation using Potential Field Methods and Bug algorithms.

---

3. Experimental Setup

• Aruco Marker-based Localization: Camera calibration ensures accurate robot position estimation.
• Camera Calibration: Eliminates distortion and improves obstacle measurement accuracy.
• ROS Node Implementation: Publishes real-world coordinates for navigation.

---

4. Dynamic Path Planning

4.1 Path Planning Algorithms Studied:

1. Dijkstra's Algorithm - Static environment optimization.
2. A star Algorithm - Heuristic-based search for efficient pathfinding.
3. D star Algorithm - Dynamic adaptation with continuous path updates.
4. D star Lite Algorithm - Simplified real-time replanning.
5. Bug2 Algorithm - Reactive boundary-following approach.

4.2 Bug-D: Hybrid Approach

• Combines Bug2's reactive avoidance with D star Lite’s real-time replanning.
• Prioritizes direct line-of-sight movement while adapting to dynamic obstacles.
• Leverages sensor fusion (LiDAR + Camera) for better environment perception.

---

5. Simulation and Testing

5.1 Gazebo Simulation

• Implemented Bug-D on TurtleBot3 in a warehouse environment.
• Navigated from Point P1 to Point P2 while avoiding dynamic obstacles.
• 

5.2 Performance Evaluation

• Path Efficiency: 1.1 (110% of the optimal path).
• Adaptability: Quick response to moving obstacles (~20 ms replanning time).
• Real-time Performance: Minimal delay (~10 ms between obstacle detection and replanning).
• Computational Efficiency: Low resource utilization for real-world feasibility.

---

6. Real-World Testing

6.1 Implementation on TurtleBot3

• Conducted tests in a dynamic lab environment with moving obstacles.
• Sensor Fusion improved localization accuracy (~0.2m error reduction in coordinates).
• Performance mirrored simulation results but revealed challenges in highly dynamic setups.

6.2 Observations & Learnings

• Real-time replanning is effective but can be improved for extreme cases.
• Sensor noise impacts localization, requiring robust fusion techniques.
• Dense obstacle environments increase computation, highlighting the need for further optimization.

---

7. Future Improvements

1. Integrating Probabilistic Motion Planning (RRT) to handle extreme obstacle movements.
2. Enhancing Sensor Fusion using Kalman Filtering or Dynamic Bayesian Networks.
3. Applying Machine Learning for real-time object recognition and classification.

---

8. Conclusion

This project successfully implemented Bug-D with sensor fusion for dynamic path planning in real-world environments. The findings demonstrate robust adaptability with potential applications in industrial automation, search & rescue, and autonomous navigation.

Further research will focus on improving performance in highly dynamic environments and advancing AI-driven decision-making for real-time robotics.

---

9. References

1. Bug2 Algorithm - Automatic Addison
2. ROS-based Path Planning
3. Dynamic Planning - Carnegie Mellon University

---