Towards heterogeneous robot team path planning in dynamic search and rescue scenarios

Roman Lavrenov, Aidar Gabdullin, Ilya Afanasyev, Vadim Indelman, Evgeni MagidThe Proceedings of JSME Annual Conference on Robotics and Mechatronics (Robomec) 2016 · DOI

Abstract

Our research focuses on the effective operation of a heterogeneous robotic group designed to carry out point-to-point navigation in GPS-denied dynamic environments. By applying a mixed local and global planning approach, alongside instruments such as planning in belief space and SLAM, we aim to enhance autonomy in search and rescue (SAR) missions. This paper introduces our heterogeneous robot team, their modeling in ROS/Gazebo, and a homotopy-based high-level planner solution.

Introduction: The Challenge of Denied Environments

Search and rescue operations frequently occur in "denied" environments where GPS signals are obstructed by structural debris and existing maps are rendered obsolete by collapses. Single-robot systems often struggle with these constraints; ground vehicles (UGVs) offer high endurance but limited visibility, while aerial vehicles (UAVs) provide superior perspectives but have limited battery life. Our work proposes a collaborative framework where a heterogeneous team shares information to navigate safely despite environmental uncertainty.

Methodology and System Architecture

Heterogeneous Hardware

The robotic team is composed of a Servosila “Engineer”—a rugged, tracked UGV equipped with a 5-DOF manipulator—and a fleet of Parrot AR.Drone 2.0 quadrotors. The "Engineer" serves as the primary ground actor, capable of traversing difficult terrain and performing manipulation tasks.

The Servosila "Engineer" UGV with its control station and VR interface.

Path Planning and Belief Space

To navigate effectively, we utilize a multi-layered planning approach:

  • Homotopy-Based Planning: Instead of a single optimal path, the planner identifies distinct classes of trajectories (e.g., left vs. right around an obstacle). Pareto optimization then selects the best class based on safety, energy, and length.
  • Belief Space Planning: Given the lack of GPS, robots account for statistical uncertainty in their positioning. They choose paths that not only reach the goal but also improve self-localization by moving near known landmarks.
  • Collaborative Mapping: UAVs act as scouts, updating global maps in real-time to inform the UGV's path.

Simulation and Modeling

A critical component of this research is the high-fidelity modeling of the robots within the ROS (Robot Operating System) and Gazebo environments. We developed detailed CAD-based models to ensure that virtual testing accurately reflects real-world physics.

High-fidelity 3D model of the "Engineer" crawler used for ROS/Gazebo simulation.

Results and Future Work

The current phase of research has successfully integrated the complex kinematics of the "Engineer" crawler into the Gazebo simulator and established the conceptual framework for the homotopy-based planner. The hardware is confirmed as SAR-ready, with the UGV offering a 4–15 hour battery life. Future efforts will focus on rigorous simulation testing against dynamic obstacles and subsequent real-world verification in controlled rubble environments.

Summary updated

Abstract

Our research focuses on effective operation of heterogeneous robotic group that could carry out point-to point navigation in GPS-denied dynamic environment, applying a mixed local and global planning approach, and such instruments as planning in belief space and SLAM. In this paper we introduce our heterogeneous group robots and their modelling in ROS/Gazebo, and briefly describe Homotopy based high level planner solution.
BibTeX
@inproceedings{lavrenov2016towards,
  author = {Roman Lavrenov and Aidar Gabdullin and Ilya Afanasyev and Vadim Indelman and Evgeni Magid},
  title = {{Towards heterogeneous robot team path planning in dynamic search and rescue scenarios}},
  booktitle = {The Proceedings of JSME Annual Conference on Robotics and Mechatronics (Robomec)},
  year = {2016},
  doi = {10.1299/jsmermd.2016.1a1-03b1},
}
Figure