RMPD: Robot motion and performance database

RMPD: Robot motion and performance database G. Barros*1, N. Appel1 and B. Kuhlenkötter1 (1) Chair of Production Systems, Ruhr-Universität Bochum, 44801 Bochum, Germany Contact information: * [e-mail] Background This dataset provides extensive motion and performance measurements from an ABB IRB 4400 industrial robot. It is intended to help researchers predict robot trajectory accuracy and analyze robot behavior using data-driven methods. Currently, the dataset features one robot platform, but future updates will add more robot models for cross-platform studies. It is developed at the Chair of Production Systems at the Ruhr-Universität Bochum. The dataset has over 30,000 recorded trajectories and more than 100,000 motion segments. Each entry includes motion data, embeddings, and performance metrics from methods like SIDTW, Euclidean distance, QDTW, and geodesic distance. These measurements show the deviations between commanded and measured TCP paths, indicating how closely the robot follows its programmed trajectory in position and orientation. All measured data was collected with a Leica AT960 laser tracker at 1000 Hz. The commanded data was streamed from the robot controller. The dataset does not include any augmented data or duplicates. The database is built to support: analysis of trajectory similarity using retrieval methods prediction of how accurately the robot moves benchmarking different methods for comparing time-series data general large-scale data-driven robotics research Dataset structure The database uses a two-schema PostgreSQL setup to organize motion data and evaluation results. It includes tables for trajectory details, segment-level motion records, embedding data, and performance results. Each trajectory consists of two or more segments, according to the robot's program structure. Performance metrics cover TCP path accuracy, measured with SIDTW and Euclidean distance, and TCP orientation accuracy, measured with QDTW and geodesic distance. Results are given as both overall statistics and point-by-point differences. All measurements use linear and circular motion types to reflect real robot behavior. The download includes a PostgreSQL database dump (.dump), a schema-only export (.sql), and sample CSV files from six trajectories for inspection without restoring the entire database. Schema Table Description motion traj_info Trajectory and measurement general information motion traj_metadata Segment-to-trajectory mapping and execution metadata motion traj_embeddings Fixed-length vector embeddings for similarity search motion traj_setpoints Commanded target points from the robot program motion traj_joint_states Commanded joint angle values during motion motion traj_position_cmd Commanded TCP position (x, y, z) motion traj_orientation_cmd Commanded TCP orientation (quaternions) motion traj_pose_act Measured TCP pose (position + orientation) motion traj_pose_act_raw Raw measured TCP pose before coord. system transformation motion traj_vel_cmd Commanded TCP velocity motion traj_vel_act Measured TCP velocity motion traj_accel_cmd Commanded TCP acceleration motion traj_accel_act Measured TCP acceleration motion robot_info Robot platform general information evaluation sidtw_info Aggregated SIDTW evaluation statistics evaluation sidtw_evaluation Point-level SIDTW deviations evaluation ed_info Aggregated Euclidean distance statistics evaluation ed_evaluation Point-level Euclidean deviations evaluation qdtw_info Aggregated QDTW evaluation statistics evaluation qdtw_evaluation Point-level QDTW deviations evaluation gd_info Aggregated geodesic distance statistics evaluation gd_evaluation Point-level geodesic deviations Database statistics 31,025 trajectories 103,073 segments Average segments per trajectory: 3.32 Average sequence length: 1,437 samples (commanded), 3,011 samples (measured) Average segment length: ~450 samples (commanded), ~942 samples (measured) Motion types: linear and circular primitives Payload range: 12 kg to 22 kg Workspace volume: 1.5 m × 2.2 m × 1.6 m Database size: approximately 150 GB Installation Install the pgvector extension: sudo apt install postgresql-16-pgvector Set up a database: CREATE DATABASE rmpd;\c rmpdCREATE SCHEMA motion;CREATE EXTENSION vector SCHEMA motion; Restore the dump: pg_restore -d rmpd -v rmpd.dump