AgriLiRa4D: A Multi-Sensor UAV Dataset for Robust SLAM in Challenging Agricultural Fields

[Website] [arXiv] [Hugging Face]

Overview

AgriLiRa4D is a novel multi-modal dataset specifically developed for agricultural UAV SLAM research, incorporating LiDAR, 4D Radar, and IMU measurements collected using an agricultural UAV platform. Our dataset features high-precision ground truth trajectories obtained from a Fiber Optic Inertial Navigation System (FINS) module with a built-in Real-Time Kinematic (RTK) receiver (FINS_RTK), ensuring centimeter-level position accuracy and high-fidelity orientation references. This work aims to advance the development of robust SLAM systems tailored for agricultural autonomous operations.

Flat Farmland

Hilly Farmland

Terraced Farmland

News

• Dec. 6, 2025 - Dataset upload is complete, and the project page as well as the preprint paper are now accessible.

Mapping Results

Scene	LiDAR	4D Radar
Flat Farmland
Hilly Farmland
Terraced Farmland

Quick Start

The dataset can be downloaded from Hugging Face.

The 33 sequences of our dataset were collected across three representative farmland terrains—flat plains, hilly regions, and mountainous terraces—located in Nanjing, China. The dataset is organized into six sequence groups based on terrain type and scanning mode (boundary or coverage), namely NJFlatB, NJFlatC, NJHillB, NJHillC, NJTerrB, and NJTerrC.

For all sequences except NJTerrB and NJTerrC, the UAV flew at a constant altitude with respect to the take-off point. In contrast, for the mountainous-terrain sequences NJTerrB and NJTerrC, the UAV maintained a fixed height Above Ground Level (AGL) to ensure flight safety and stable sensor coverage over rapidly varying elevation. Multiple combinations of flight altitudes and speeds were employed to introduce different levels of SLAM difficulty. Each sequence additionally begins with a short stationary or hovering segment to facilitate IMU initialization.

Scene	Sequence	Scanning Mode	Altitude (m)	Speed (m/s)	Path Length (m)
Flat Farmland	NJFlatB01	boundary	5	3	434.77
	NJFlatB02	boundary	5	8	464.21
	NJFlatB03	boundary	10	3	456.32
	NJFlatB04	boundary	10	8	462.18
	NJFlatB05	boundary	15	3	465.89
	NJFlatB06	boundary	15	8	454.21
	NJFlatC01	coverage	5	8	805.65
	NJFlatC02	coverage	10	3	801.17
	NJFlatC03	coverage	10	8	798.96
	NJFlatC04	coverage	15	3	822.23
Hilly Farmland	NJHillB01	boundary	8	3	490.61
	NJHillB02	boundary	8	8	493.07
	NJHillB03	boundary	13	3	480.98
	NJHillB04	boundary	13	8	484.60
	NJHillB05	boundary	18	3	483.84
	NJHillB06	boundary	18	8	488.41
	NJHillC01	coverage	8	3	776.47
	NJHillC02	coverage	8	8	783.31
	NJHillC03	coverage	13	3	761.55
	NJHillC04	coverage	13	8	768.14
	NJHillC05	coverage	18	3	756.07
	NJHillC06	coverage	18	8	769.94
Terraced Farmland	NJTerrB01	boundary	3	3	204.91
	NJTerrB02	boundary	6	3	207.21
	NJTerrB03	boundary	6	6	209.71
	NJTerrB04	boundary	9	3	211.95
	NJTerrB05	boundary	9	6	215.72
	NJTerrC01	coverage	3	3	311.23
	NJTerrC02	coverage	3	6	307.53
	NJTerrC03	coverage	6	3	311.24
	NJTerrC04	coverage	6	6	300.84
	NJTerrC05	coverage	9	3	313.64
	NJTerrC06	coverage	9	6	317.48

The Radar-LiDAR-Inertial sensor data is provided in ROS bag format. Each bag contains the following topics:

Sensor	Module	Topic Name	Message Type	Rate (Hz)
LiDAR	Robosense Airy	/rslidar_points	sensor_msgs/PointCloud2	10
IMU	Built-in (LiDAR)	/rslidar_imu_data	sensor_msgs/IMU	200
4D Radar	Mindcruise A1	/radar_points	sensor_msgs/PointCloud2	10
FINS_RTK	TJ-FINS70D	/aircraft_pose_enu	geometry_msgs/PoseStamped	100
		/aircraft_pose_flu	geometry_msgs/PoseStamped	100
		/aircraft_position_llh	sensor_msgs/NavSatFix	100

For /rslidar_points, pay attention to its point's timestamp: the per-point timestamp has a constant offset w.r.t. the message header.stamp, and its unit is seconds (s).

For /rslidar_points and /radar_points, use custom PCL PointT type:

namespace robosense {
struct EIGEN_ALIGN16 Point {
  PCL_ADD_POINT4D;
  float intensity;
  std::uint16_t ring = 0;
  double timestamp = 0;
  EIGEN_MAKE_ALIGNED_OPERATOR_NEW
};
} // namespace robosense
// clang-format off
POINT_CLOUD_REGISTER_POINT_STRUCT(robosense::Point,
    (float, x, x)
    (float, y, y)
    (float, z, z)
    (float, intensity, intensity)
    (std::uint16_t, ring, ring)
    (double, timestamp, timestamp)
)

// clang-format on
namespace txg_radar
{
struct EIGEN_ALIGN16 Point
{
  PCL_ADD_POINT4D;      // preferred way of adding a XYZ+padding
  float v_doppler_mps;  // Doppler velocity in m/s
  float snr_db;         // Signal-to-noise ratio in dB
  float rcs;            // Radar cross-section in m^2
  EIGEN_MAKE_ALIGNED_OPERATOR_NEW
};
}  // namespace txg_radar
// clang-format off
POINT_CLOUD_REGISTER_POINT_STRUCT(txg_radar::Point,
    (float, x, x)
    (float, y, y)
    (float, z, z)
    (float, v_doppler_mps, v_doppler_mps)
    (float, snr_db, snr_db)
    (float, rcs, rcs)
)

Ground Truth

One of the three ground-truth representations (FLU reference) is provided using TUM trajectory format in this repository, along with the sensor extrinsic parameters of this acquisition device.

Note: When performing ground-truth analysis, the estimated state must first be gravity-aligned, and then the IMU's odometry must be transformed to the body's odometry based on an FLU reference!!!

Here is an example about how to convert gravity-aligned IMU's odometry obtained from a LIO/RIO/RLIO system into body's odometry based on an FLU reference.

// FLU to gravity-aligned odom
M3D R_flu_odom;
R_flu_odom << 0, 1, 0, 
              -1, 0, 0, 
              0, 0, 1;
V3D t_flu_odom(0.0, 0.0, 0.0);

// body to imu
M3D R_airbody_imu;
R_airbody_imu << 0, 0, -1, 
                1, 0, 0, 
                0, -1, 0;
V3D t_airbody_imu(0.0, 0.0, 0.0);

// We can get `R_odom_imu` and `t_odom_imu` through LIO/RIO/RLIO (gravity-aligned)
M3D R_flu_imu = R_flu_odom * R_odom_imu;
V3D t_flu_imu = R_flu_odom * t_odom_imu + t_flu_odom;
M3D R_flu_body = R_flu_imu * R_airbody_imu.transpose();
V3D t_flu_body = t_flu_imu - R_flu_body * t_airbody_imu;

Related Work

1. FAST-LIO2: Fast Direct LiDAR-inertial Odometry
2. Faster-LIO: Lightweight Tightly Coupled Lidar-inertial Odometry using Parallel Sparse Incremental Voxels
3. EKF-RIO: Radar Inertial Odometry With Online Calibration
4. GaRLIO: Gravity enhanced Radar-LiDAR-Inertial Odometry

Citation

@misc{zhan2025agrilira4dmultisensoruavdataset,
      title={AgriLiRa4D: A Multi-Sensor UAV Dataset for Robust SLAM in Challenging Agricultural Fields}, 
      author={Zhihao Zhan and Yuhang Ming and Shaobin Li and Jie Yuan},
      year={2025},
      eprint={2512.01753},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2512.01753}, 
}