All Projects
3rd Place Global Published

ICRA 2023 BARN Challenge — 3rd Place Global

Autonomous mobile robot navigation in highly constrained, obstacle-dense environments — ranked 3rd globally in both simulation and real-world rounds

ROS ROS2 C++ Python Autonomous Navigation Path Planning A* Algorithm Gazebo Simulation Navigation Stack Parameter Optimization

Overview

Achieved 3rd place globally in both the simulation and real-world rounds of the ICRA 2023 BARN Challenge (Benchmark Autonomous Robot Navigation). The BARN challenge tests autonomous navigation in highly constrained environments — narrow corridors, cluttered spaces, and dense obstacle configurations that cause conventional navigation stacks to fail.

Competing against teams from leading robotics institutions worldwide, the system successfully demonstrated robust navigation performance across all BARN benchmark environment categories, culminating in a published paper in the ICRA 2023 proceedings.

The Challenge

The BARN benchmark is specifically designed to find the failure modes of autonomous navigation systems. Environments include extremely narrow passages (less than robot-width clearance), unpredictable obstacle layouts, and dead-end configurations. Many standard planners either fail to find paths or produce unsafe trajectories.

The real-world round added hardware variability, sensor noise, and physical obstacles that simulation cannot fully replicate. The system had to perform reliably without any fine-tuning between the simulated and physical environments — a true test of generalization.

Technical Approach

  • Full autonomous navigation stack designed and tuned using ROS, C++, and Python for constraint-dense environments
  • Local and global planner optimization with custom configurations targeting the BARN environment distribution's failure modes
  • Parameter optimization methodology developed specifically to generalize across the full BARN environment distribution rather than overfitting to individual maps
  • Extensive simulation testing in Gazebo across all BARN obstacle configuration categories including narrow corridors and cluttered layouts
  • Real-world validation with a physical robot in cluttered test environments matching BARN specifications
  • Custom recovery behaviors implemented for dead-end detection and narrow-passage negotiation scenarios
  • 75% success rate achieved in real-world obstacle-dense environments — a significant result given the difficulty of the BARN benchmark

Key Outcomes

3rd Global Simulation Round
3rd Global Real-world Round
75% Real-world Environments
ICRA 2023 Conference Paper

Published in the ICRA 2023 proceedings: "Autonomous Ground Navigation in Highly Constrained Spaces: Lessons learned from the 2nd BARN Challenge at ICRA 2023"