English | 简体中文
Current Status: ✅ v1.2.0 Completed — FUEL Autonomous Exploration Reproduced on the Real Platform Latest Version: v1.2.0
This repository documents the iterative development of a custom 250mm wheelbase quadrotor designed for advanced autonomous navigation and control research.
The goal is to build a robust hardware and software foundation, starting from classical SLAM and planning algorithms, and gradually migrating towards end-to-end / learning-based perception and control systems.
| FUEL — Frontier-based Autonomous Exploration in an unknown indoor corridor |
|---|
 |
| Full video |
| FAST-LIO2 Localization — Stable Hover | DLIO + EGO-Planner — Autonomous Waypoint Navigation (4× speed) |
|---|---|
 |
 |
| Full video | Full video |
The drone is built on a custom 250mm carbon fiber frame, featuring a high-performance compute module for onboard AI processing.
| Component | Model / Specs | Note |
|---|---|---|
| Onboard Computer | NVIDIA Jetson Orin NX (8GB) | Core computing unit |
| Flight Controller | Holybro Pixhawk 4 | 32-bit, running PX4/ArduPilot |
| LiDAR | Livox Mid-360 | Omnidirectional 3D LiDAR |
| Depth Camera | Intel RealSense D435 | For VIO and depth sensing |
| Frame | Custom Carbon Fiber (250mm) | "Taobao" Custom Cut |
| Motors | Koofly 2207 V2 (1960KV) | 4S Power System |
| ESC | EMAX BLHeli32 (45A) | 4-in-1 ESC supported |
| Battery | Gens Ace 4S 4000mAh | Long endurance for computing |
| Propellers | Gemfan 5147 (3-blade) | Efficient 5-inch props |
| RC System | RadioLink AT9S Pro + R12DSM | Reliable control link |
📄 For a complete list of parts, cables, and tools, please check the BOM List.
graph LR
A[Livox Mid-360] -->|/livox/lidar, /livox/imu| B[livox_ros_driver2]
B --> C["FAST-LIO2 / DLIO<br>(LiDAR-Inertial Odometry)"]
C -->|"/Odometry_highrate (~IMU rate)"| D[px4ctrl]
C -->|/Odometry + /cloud_registered| P{Planner}
P -->|EGO-Planner| E[Waypoint Navigation]
P -->|FUEL| F[Autonomous Exploration]
E -->|position commands| D
F -->|position commands| D
D --> M[MAVROS] --> G[PX4 / Pixhawk 4]
The stack is organized by role — each layer is an independent catkin workspace:
- Localization (
Drone/localization/) — FAST-LIO2 or DLIO, both publishing/Odometry+/Odometry_highrate+/cloud_registered. - Planner (
Drone/planner/) —ego_planner(point-to-point / waypoint navigation) andFUEL(frontier-based autonomous exploration). Start withscripts/run_fastlio_ego_px4.sh/scripts/run_dlio_ego_px4.sh. - Control (
Drone/control/) —px4ctrl(+uav_utils,quadrotor_msgs,cmake_utils) bridging position commands to PX4 via MAVROS (MAVROS itself is apt-installed). - VIO (
Drone/vio/) — VINS-Fusion, optional visual-inertial front-end (not used by the LiDAR flight stack; kept for future vision work).
| Module | Upstream (pinned commit) | What was changed here |
|---|---|---|
FAST-LIO2 (Drone/localization/fast_lio2_ws) |
hku-mars/FAST_LIO 7cc4175 |
① High-rate odometry: new PropagateHighRateState() forward-propagates the latest LIO state with each incoming IMU sample and publishes /Odometry_highrate (~IMU rate) for px4ctrl, instead of the ~10 Hz LiDAR-rate /Odometry. ② Ported from livox_ros_driver to livox_ros_driver2. ③ Mid-360 config tuning (config/mid360.yaml). |
DLIO (Drone/localization/dlio_ws) |
vectr-ucla/direct_lidar_inertial_odometry fc8d183 |
① New imu/accelScale parameter — the Mid-360 IMU reports acceleration in g, not m/s², so it is rescaled by 9.80665. ② Mid-360 extrinsics in cfg/dlio.yaml (aligned with FAST-LIO2's mid360.yaml). ③ New launch/dlio_mid360_drone.launch remaps DLIO outputs to /Odometry_highrate + /cloud_registered, making DLIO a drop-in replacement for FAST-LIO2 in the same stack. |
EGO-Planner (Drone/planner/ego_planner, from Fast-Drone-250 8ff6427) |
launch/config | ① Odometry source switched from VINS-Fusion (/vins_fusion/imu_propagate) to LIO (/Odometry). ② grid_map is built directly from the LiDAR registered cloud (cloud_registered) instead of the depth camera. ③ max_acc lowered 6.0 → 2.0 for safe indoor flight. Full diff in patches/. |
FUEL (Drone/planner/FUEL) |
HKUST-Aerial-Robotics/FUEL 662dd23 |
Frontier-based autonomous exploration reproduced on the real platform: takes the Mid-360 LiDAR cloud + LIO odometry as input and drives exploration through exploration_real.launch. Reproduced and verified in a real indoor corridor (see demo). |
px4ctrl (Drone/control/px4ctrl, from Fast-Drone-250 8ff6427) |
launch/config | Position-control bridge to PX4. Driven by /Odometry_highrate; run_ctrl.launch + ctrl_param_fpv.yaml tuned for the 250mm platform. Shipped with its deps uav_utils, quadrotor_msgs, cmake_utils. |
VINS-Fusion (Drone/vio/VINS-Fusion) |
HKUST-Aerial-Robotics/VINS-Fusion | loop_fusion OpenCV-4.2/4.5 link fix (see patches/vins_loop_fusion_opencv4.2_fix.patch and Drone/vio/README.md). Optional VIO front-end, not used by the LiDAR flight stack. |
| livox_ros_driver2 | Livox-SDK/livox_ros_driver2 6b9356c |
Not vendored — use upstream directly. Only the network config differs: see config/livox/MID360_config.json (host 192.168.1.50, LiDAR 192.168.1.154). |
patches/contains the exact diffs of the modified modules against their pinned upstream commits, for review and re-application.
.
├── README.md / README.zh-CN.md # This file (EN) / Chinese version
├── BOM.xlsx # Detailed hardware Bill of Materials
├── docs/assets/ # Demo GIFs embedded above
├── videos/ # Full-length demo videos (mp4)
├── Drone/
│ ├── localization/
│ │ ├── fast_lio2_ws/ # Modified FAST-LIO2 (high-rate odom, driver2 port)
│ │ └── dlio_ws/ # Modified DLIO (accelScale, Mid-360 launch)
│ ├── planner/
│ │ ├── ego_planner/ # EGO-Planner extracted from Fast-Drone-250 (waypoint nav)
│ │ └── FUEL/ # FUEL frontier-based autonomous exploration
│ ├── control/ # px4ctrl + uav_utils + quadrotor_msgs + cmake_utils
│ └── vio/ # VINS-Fusion (optional VIO; loop_fusion OpenCV fix)
├── config/livox/ # Mid-360 network config for livox_ros_driver2
├── patches/ # Diffs vs. pinned upstream commits
└── scripts/ # One-shot startup / takeoff / record scripts
# 1. Build the Livox driver (upstream) with the config from this repo
# livox_ws/src/livox_ros_driver2 @ 6b9356c, replace config/MID360_config.json
# 2. Build the workspaces from this repo's sources (each is an independent catkin_make ws)
# Drone/localization/fast_lio2_ws Drone/localization/dlio_ws
# Drone/control (px4ctrl + deps) Drone/planner/ego_planner Drone/planner/FUEL
# 3a. Waypoint navigation: launch the full stack (LIO + relay + MAVROS + px4ctrl + EGO-Planner)
./scripts/run_fastlio_ego_px4.sh # FAST-LIO2 + EGO-Planner pipeline
./scripts/run_dlio_ego_px4.sh # or: DLIO + EGO-Planner pipeline
# 3b. Autonomous exploration: build Drone/planner/FUEL, then
roslaunch exploration_manager exploration_real.launch
# 4. After all topics are healthy, take off
./scripts/takeoff.shFocus: Establishing a stable flight platform using Visual-Inertial Odometry (VIO).
- Localization: VINS-Fusion (Stereo + IMU)
- Planning: EGO-Planner
- Key Hardware: Jetson NX + D435
Focus: Integrating Livox Mid-360 for robust, high-precision localization in complex environments.
- Localization: FAST-LIO2 / DLIO (LiDAR-Inertial Odometry) with Livox Mid-360
- Planning: EGO-Planner for waypoint (fixed-point) navigation
- Result: Stable indoor localization (hover) and autonomous waypoint navigation verified on the real platform.
Focus: From waypoint navigation to fully autonomous exploration of unknown environments.
- Autonomous Exploration: FUEL (Fast UAV ExpLoration) reproduced on the real platform, driven by Mid-360 LiDAR + LIO odometry.
- Result: Frontier-based autonomous exploration verified in a real indoor corridor — see the demo GIF above.
Focus: Move from modular SLAM + planning toward learning-based, end-to-end navigation on the real drone.
- Deploy an end-to-end navigation policy (learning-based perception → action) onboard the Jetson.
- Sim-to-Real transfer for the navigation policy.
- Integration of neural perception modules with the existing LiDAR-inertial localization stack.
This project builds on the excellent open-source work of hku-mars/FAST_LIO, vectr-ucla/direct_lidar_inertial_odometry, ZJU-FAST-Lab/Fast-Drone-250 (EGO-Planner), HKUST-Aerial-Robotics/FUEL, and Livox-SDK/livox_ros_driver2.