Front-Facing LiDAR Altitude Estimation
Type of Project: Commercial Internship at Perciv AI, Real-Time Robotics
Status: Commercially implemented in ABZ Innovation agricultural drones
Summary:
Developed, tested, and deployed a robust algorithm enabling agricultural drones to estimate their altitude using only a front-facing LiDAR sensor—supporting reliable low-altitude flight without GPS or downward sensors.
- Fewer sensors reduce cost and risk of hardware failure
- Can be used as a primary or backup altitude estimator for greater robustness
- More accurate than GPS in complex and irregular environments
Technical Stack
- Languages & Tools: C++, Python, ROS, Docker, GitHub, Linux
- Embedded Platforms: Raspberry Pi 5, medium-sized UAVs
- Sensor & Robotics Skills: Point cloud processing, sensor calibration, coordinate frame transformations (TF), real-time filtering, hardware-software integration, hardware-in-the-loop (HIL) testing, system latency optimization
- Software Engineering: OOP, modular architecture, unit testing (C++ & Python), Git-based workflows, robust documentation, scalable codebase design
- Simulation & Visualization: Foxglove Studio, RViz
- Soft Skills & Collaboration: Technical writing, team-wide safety training, live demonstrations, documentation for handover and future development
These images are the propery of PERCIV AI, do not use or share them without consent
Live field testing in an unstructured environment to validate altitude estimation accuracy
Real-time point cloud processing demonstration
Flight test with the algorithm running on-board
Embedded module executing the estimator during field trials
Key Contributions
- Algorithm Development: Designed real-time altitude estimation algorithms in Python and C++ using ROS, with seamless integration in Linux and Docker environments.
- Embedded Systems Integration: Deployed code on Raspberry Pi 5 modules and physically integrated into drone hardware for field use.
- Real-World Testing: Performed live drone tests in agricultural settings, gathering sensor data and iterating on code for improved performance.
- Hardware-in-the-Loop (HIL): Built HIL setups to simulate sensor input for optimizing latency and compute load in constrained environments.
- System Design & Knowledge Transfer: Implemented modular ROS node architecture, led a functional safety lecture for the engineering team, and documented the full system for future scaling and maintenance.