LiDAR odometry · Power & sensing stack

LiDAR SLAM Upgrade

2024 · Post-thesis

Continuation of my Matura thesis drone: Pixhawk 4 + Jetson autonomy stack, rebuilt for LiDAR odometry instead of GPS. The thesis LD06 was replaced by a Unitree L1 on a Jetson companion computer running the SLAM/odometry pipeline, while PX4 still handles attitude and low-level control.

A custom multi-rail power PCB (the stock power distribution board could not provide the voltages and currents needed to feed Jetson + LiDAR), mechanical work to kill vibration on the spinning sensor (counterweight, TPU mount, prop guards had to go because they resonated with the LiDAR), inverted props for efficiency gain, and frame weight cuts so the aircraft didn't become even heavier. Despite the heavier sensor, the upgraded build weighs ~2.45 kg—only about 20 g more than the thesis version.

Upgraded drone with Unitree L1 LiDAR, Jetson stack, and custom power PCB on a scale reading 2448 g — Full stack
Everything on one frame: the Unitree L1, Jetson companion computer, custom power PCB, and the original Pixhawk stack. Weighed at ~2.45 kg despite the addition of the heavy LiDAR it only weighs 20g more than the original thesis version. This is thanks to weight cuts on the frame (removing folding joints for arms, removing landing gear, 3D Printing more weight efficient parts.).

Test flight

Indoor hop with the upgraded perception and power stack—first stable indoor flight with LiDAR odometry.

PCB layout in EDA tool — multi-rail power distribution with VCC and GND pours — PCB layout
Layout in the EDA tool: separate rails for Jetson and LiDAR peripherals, thick traces to handle full 3A for jetson USB-C port

3D render of custom multi-rail power distribution PCB — 3D render

Assembled custom power distribution PCB with capacitors, inductor, USB-C, and XT30 connector — Assembled board

Unitree L1 LiDAR opened — internal rotor and PCB visible — LiDAR
Opened after the sensor broke was curious how the spinning parts looked like and why causes such high vibration. Unitree sent a free warranty replacement :)

Brushless motor with downward-facing propeller on carbon arm — Propulsion
I tried prop guards, but they resonated with the LiDAR vibration frequency. I also inverted the props to face downward: a significant free efficiency gain.

SLAM

I adapted FAST-LIO to work with the Unitree L1 LiDAR, and also tried Unitree’s Point-LIO adaptation for the L1. FAST-LIO had fewer moments where the estimate completely drifted, but Point-LIO required much less compute.

LiDAR rotor balancing

The L1’s spinning mass was badly unbalanced—the vibration didn’t just distort scans, it made the entire drone unstable in flight. I added counterweight on the rotor to correct the imbalance; a photo of the 3D-printed balancing mass will be added here later.