A Multimodal Motion Planning Dataset: Synchronized Real-World and Simulation Recordings of a UR10e in a Bus Chassis Demonstrator

This dataset provides a comprehensive, synchronized collection of multimodal sensor data captured from a Universal Robots UR10e manipulator mounted on a linear axis, operating within a highly constrained industrial environment (a 4-compartment bus chassis demonstrator).

The primary value of this dataset lies in its millisecond-level synchronization between the Physical Robot and its Digital Twin (Gazebo Simulation), enabled by a custom ROS 2 (Humble), Apache Kafka, and Elasticsearch data pipeline. It is designed to support research in motion planning benchmarking, sim-to-real transfer, and robotic process automation.

Key Features:

Algorithms: Includes trajectory executions from 11 distinct OMPL planners (e.g., RRTConnect, SBL, PRMstar, KPIECE, EST, LBKPIECE) under identical start/goal configurations.

Volume: Data covers 110 experimental runs (10 repetitions per algorithm).

Multimodal Data: Contains high-frequency (50 Hz) recordings of:

Joint States: Position, velocity, and effort for 6 DoF arm + 1 DoF linear axis.

Dynamics: Raw Force/Torque sensor readings (Wrench) from the TCP.

Kinematics: Real-time Cartesian TCP pose (position and orientation).

Visuals: External camera recordings (.avi) and simulation screencasts (.webm).

Planning Data: Planned trajectories in JSON format for both real and sim environments.

Dataset Organization:
The repository is structured into two main directories:

Data in Ready (Processed): Consolidated CSV files containing time-series data from ROS topics (Joint States, TCP Pose, Force/Torque), synchronized and cleaned for immediate analysis.

Trajectories (Raw & Visual): Sub-divided folders for each motion planning algorithm. Each folder contains the specific real_trajectories.json, sim_trajectories.json, raw controller logs, and video documentation of the experiments.

Steps to Reproduce / Methodology:
The data was collected using a "Hardware-in-the-Loop" (HIL) architecture. A ROS 2-based controller executed the motion plans generated by MoveIt2/OMPL on both the physical UR10e and the Gazebo simulation simultaneously. Data streams were buffered via Apache Kafka to ensure zero data loss and synchronized using NTP-aligned ROS header timestamps before being exported to CSV and JSON formats.