GeoScout is a fully autonomous, multi-sensor robotic platform engineered for LiDAR-based surface reconstruction and Ground Penetrating Radar (GPR)-assisted subsurface characterization in GPS-denied, high-risk underground mining environments. The system integrates a heterogeneous computing architecture—Raspberry Pi 5 for high-level ROS 2 middleware execution and Arduino Mega for deterministic, low-level motor/sensor control—coordinated over ROS 2 pub/sub interfaces.
Platform Architecture
Mechanical Subsystem:
A custom 6-wheel differential drive chassis employing a modified rocker–bogie suspension enables passive terrain adaptation for highly irregular mine floors, ensuring continuous wheel–ground contact and stability during sensor acquisition.
Perception Stack:
- RPLIDAR A1M8 provides 360° 2D ranging for SLAM and occupancy grid generation.
- Kinect depth camera generates dense 3D point clouds to augment LiDAR-based SLAM through depth fusion.
- A custom GPR module uses an AD9833 DDS chirp source, TL072 conditioning amplifier, and TDA7294 high-power stage to interrogate subsurface strata; reflected waveforms are digitized, timestamped, and streamed to ROS for real-time visualization and anomaly inference.
Control & Computation:
- Raspberry Pi 5 executes the SLAM pipeline, Nav2 path planning/behavior-tree navigation, and multi-sensor fusion.
- Arduino Mega manages encoder-driven PID motor control and high-speed GPR timing signals.
- ROS 2 Jazzy orchestrates all inter-node communication under a distributed, asynchronous message-passing framework.
Simulation & Validation Workflow
- Gazebo Harmonic models the rover’s multibody dynamics and sensor physics for virtual SLAM and Nav2 evaluation.
- RViz2 renders LiDAR/Kinect fused maps and GPR subsurface plots.
- MATLAB models chirp propagation, attenuation, and subsurface reflection signatures.
- Proteus validates electronic subsystems including signal generation/amplification chains.
Core Capabilities
- Autonomous navigation with ROS 2 Nav2 and differential-drive kinematics.
- Real-time SLAM via LiDAR/depth fusion.
- Subsurface anomaly detection using active GPR waveform analysis (phase/amplitude shift extraction).
- High-torque actuation with BTS7960 drivers and encoder feedback.
- Simulation-first development enabling rapid prototyping and model-to-hardware consistency.
Future Development
Planned upgrades include AI-driven motion planning, multi-modal environmental diagnostics (CH₄/CO₂/thermal), LoRa/Wi-Fi 6 telemetry, deep-learning SLAM pipelines, and multi-agent cooperative exploration for large-scale mine mapping.Github_Link
