IMU and Servomotor Signal Dataset for Indoor Surface Classification

This dataset contains inertial measurement unit signals (Xsens MTi-3 IMU) and servomotor data (Dynamixel XL430-W250) collected from a small differential-drive mobile robot operating on multiple indoor surfaces. The dataset was created to support research on terrain classification and adaptive control for mobile robots.
Experiments were conducted across 10 surfaces: laminate flooring, short carpet (6 mm pile), long carpet (15 mm pile), artificial grass, PVC foamboard, linoleum, ceramic tiles, OSB, foam underlayment (2 mm thick), and EVA foam tiles. The robot was used in two different configurations, a four-wheeled (4W) and a six-wheeled (6W) version. All runs were performed along an 8-shaped path consisting of two circles with a diameter of 0.4 m each. The trajectory execution was automated: a control script commanded the robot motion in real time, using feedback from a Vicon motion capture system to ensure precise tracking of the predefined path. During data collection, the robot’s onboard IMU recorded three-axis linear acceleration and angular velocity, while servomotor sensors provided angular velocity and estimated load torque, which can be related to power consumption. The dataset contains 669 runs, each one recorded on a single surface. Of these, there are 349 runs performed using the 4W robot configuration and 320 using the 6W configuration.
This dataset was used to demonstrate that incorporating estimated power consumption features from servomotors can improve IMU-based classification accuracy. The data is suitable for applications such as terrain-aware navigation, adaptive control, slip detection, and energy-efficient path planning.
The corresponding code (data processing and model training) can be found in the paper respository: https://github.com/lukasz-kedzierski/surface-classification