AeroNav-SLAM tackles autonomous mobile robot localization and high-density topological mapping in environments where satellite GPS signals are unavailable or corrupted by multi-path reflections. Combining a 16-channel solid-state 3D LiDAR with a 9-DoF industrial IMU, the system utilizes continuous-time factor graph optimization to generate real-time 3D voxel maps and drift-free trajectory estimates across tunnels, industrial boiler rooms, and subterranean mines.
In subterranean mines, warehouse interiors, and nuclear decommissioning zones, GPS is completely blocked. Traditional visual SLAM degrades catastrophically in low-light, dust-filled, or geometrically degenerate conditions (such as featureless smooth concrete corridors). Without drift-free localization, autonomous robotic inspections risk collision, tether loss, and permanent hardware entrapment.
Implement a tightly-coupled iterated Extended Kalman Filter (iEKF) fusing raw point-cloud scans with high-rate 200 Hz inertial measurements.
Design a dynamic Scan-Context loop closure detection algorithm capable of identifying previously visited intersections in rotation-invariant space.
Develop an adaptive voxel-grid downsampling pipeline operating in sub-30ms execution times on embedded compute.
Demonstrate autonomous waypoint navigation and dense 3D mesh generation on a ruggedized tracked mobile robot platform.
The navigation pipeline executes in two synchronized threads:
- Sensor Array: Livox Mid-360 3D LiDAR (360° horizontal FoV, 40m range, 200k pts/sec) paired with an isolated Bosch BMI088 industrial IMU.
- Processing Hardware: NVIDIA Jetson Orin Nano (40W mode, 40 TOPS AI compute) running Ubuntu 22.04 LTS and ROS2 Iron.
- Drive Subsystem: Dual brushless DC planetary-geared tracked chassis with CAN-bus motor controllers and optical quadrature wheel encoders.
- Safety Layer: Dual ultrasonic rangefinders and hardware emergency watchdog circuit with auto-braking.
- Benchmark Datasets: Validated against the Newer College and SubT subterranean benchmark datasets.
- Physical Field Testing: Deployed across 1.4 kilometers of subterranean utility tunnels and multi-level parking structures.
- Evaluation Metric: Absolute Trajectory Error (ATE) and Relative Pose Error (RPE) verified against external Leica Total Station survey ground truth.
The system runs at 10 Hz for point-cloud updates and 200 Hz for state propagation. Real-time costmaps are piped directly into a 3D TEB (Timed Elastic Band) local planner, enabling the robot to dynamically navigate around moving obstacles and unmapped pipes.
Field evaluations tested mapping stability under abrupt illumination drops, airborne particulate injection (fog/smoke machine), and severe mechanical shock over rough rocky terrain. Trajectory drift percentage per distance traveled was recorded.
- Absolute Trajectory Error (ATE): < 0.62% of total distance traveled without loop closure; < 0.08% with loop closure enabled.
- Point Cloud Density: > 1,200 points/m³ at 15m scanning distance.
- Loop Closure Convergence Time: < 18ms per query across a 5,000-keyframe library.
- Maximum Autonomous Transit Speed: 1.8 m/s in completely unlit subterranean corridors.
- Hardware: Livox Mid-360 LiDAR, BMI088 IMU, NVIDIA Jetson Orin, Tracked Robot Chassis.
- Software: C++20, ROS2, GTSAM (Georgia Tech Smoothing and Mapping), PCL (Point Cloud Library), OctoMap, Nav2 Navigation Stack.
Complete open-source C++ ROS2 LIO package optimized for ARM64 architectures.
Calibrated subterranean 3D point cloud dataset with ground truth survey benchmarks.
Hardware mounting CAD files with integrated rubber shock-isolation grommets.
Comprehensive technical report detailing covariance tuning and factor graph configuration.
Autonomous subterranean mining survey, underground sewer and tunnel infrastructure inspection, indoor disaster search-and-rescue, and GPS-denied automated guided vehicles (AGVs) in fulfillment centers.
Testing in active industrial tunnels incorporated strict safety tethering, perimeter exclusion zones, and remote RF kill-switches complying with ISO 3691-4 industrial mobile robot safety mandates.
A production-ready autonomous mobile navigation platform delivering sub-decimeter positioning accuracy in complete darkness without requiring external beacons or satellite signals.
Fusing thermal imaging cameras to detect electrical hotspots during autonomous mapping, and expanding from ground platforms to micro-aerial multi-rotor drones for confined cave exploration.
Drift-free sub-centimeter autonomous robotic localization across GPS-denied subterranean and industrial landscapes.
An autonomous mobile robotics platform utilizing tightly-coupled LiDAR-inertial odometry and factor-graph loop closures to achieve sub-centimeter 3D mapping and navigation in GPS-denied tunnels and industrial facilities.
