Enhancing Online Pose Estimation via Monte Carlo-Assisted EKF Odometry and SLAM: Collected Dataset

This dataset contains navigation and localization data for two robotic platforms (Panther and Spot) operating across different indoor environments. The data is stored in a structured format, allowing for systematic storage, retrieval, and analysis across different platforms, environments, algorithms, and iterations.
The dataset includes pose estimates from the selected SLAM algorithms along with independent AMCL and EKF pose estimates, raw and formatted occupancy grid maps, and sensor data (LiDAR and IMU) collected during the navigation tasks.

The primary objective of this dataset is to evaluate the performance of 2D SLAM algorithms with and without AMCL enhancement across different environments and robotic platforms. By providing detailed pose estimates, occupancy grid maps, and sensor data, this dataset enables researchers to analyze the impact of AMCL on SLAM performance, compare results across different algorithms, and explore the influence of environmental factors on navigation accuracy. The dataset serves as a valuable resource for advancing research in robotic navigation, localization, and mapping in structured indoor environments.

The experiment was an indoor site inspection conducted in the main AGH University of Kraków Building A0. We used the quadruped robot Spot from Boston Dynamics, equipped with an Ouster 32 3D LiDAR and an integrated 6-DOF IMU connected to an NVIDIA embedded computing unit. The system is based on ROS2, and the devices are time-synchronized via a hardware clock to ensure consistent LiDAR/IMU fusion. To ensure measurement repeatability and reduce human interference, we defined a mission in Spot’s native software by Boston Dynamics. First, we defined the robot’s path by placing dedicated QR tags, then we performed manually assisted exploration along the tags and recorded a global path to follow. Next, we switched to fully autonomous operation and repeated
the entire route twice while recording raw sensor data. The collected measurements were divided into three levels; with two iterations each, this resulted in six independent measurements for evaluation. For each of the six measurements, we ran the selected SLAM algorithms with and without AMCL feedback 10 times each, resulting in a total of 120 runs for analysis.

Research tools used: Panther robot, Spot robot, Ouster 32 3D LiDAR, ROS2 Jazzy,
Nvidia Jetson Orin, ROS2, Python 3.10 & custom data collection scripts.