Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

The Mini Wheelbot Dataset: High-Fidelity Data for Robot Learning

Henrik Hose, Paul Brunzema, Devdutt Subhasish, Sebastian Trimpe

TL;DR

This work addresses the data bottleneck in learning-based control for fast, unstable robots by introducing a high-fidelity, open-source Mini Wheelbot dynamics dataset with 1 kHz synchronized onboard and ground-truth data, plus third-person video. It catalogs diverse experiments across multiple hardware instances and surfaces using PRBS, nonlinear MPC, and RL, enabling robust dynamics learning, state estimation benchmarking, and time-series classification. A key contribution is the explicit data schema, rich metadata, and example pipelines (dynamics modeling via an autoregressive MLP, estimator benchmarking, and a transformer-based classifier) that demonstrate practical use cases. The dataset aims to democratize robotics research by providing reproducible benchmarks and plans to extend with LiDAR and vision modalities to broaden applicability across perception–control pipelines.

Abstract

The development of robust learning-based control algorithms for unstable systems requires high-quality, real-world data, yet access to specialized robotic hardware remains a significant barrier for many researchers. This paper introduces a comprehensive dynamics dataset for the Mini Wheelbot, an open-source, quasi-symmetric balancing reaction wheel unicycle. The dataset provides 1 kHz synchronized data encompassing all onboard sensor readings, state estimates, ground-truth poses from a motion capture system, and third-person video logs. To ensure data diversity, we include experiments across multiple hardware instances and surfaces using various control paradigms, including pseudo-random binary excitation, nonlinear model predictive control, and reinforcement learning agents. We include several example applications in dynamics model learning, state estimation, and time-series classification to illustrate common robotics algorithms that can be benchmarked on our dataset.

The Mini Wheelbot Dataset: High-Fidelity Data for Robot Learning

TL;DR

This work addresses the data bottleneck in learning-based control for fast, unstable robots by introducing a high-fidelity, open-source Mini Wheelbot dynamics dataset with 1 kHz synchronized onboard and ground-truth data, plus third-person video. It catalogs diverse experiments across multiple hardware instances and surfaces using PRBS, nonlinear MPC, and RL, enabling robust dynamics learning, state estimation benchmarking, and time-series classification. A key contribution is the explicit data schema, rich metadata, and example pipelines (dynamics modeling via an autoregressive MLP, estimator benchmarking, and a transformer-based classifier) that demonstrate practical use cases. The dataset aims to democratize robotics research by providing reproducible benchmarks and plans to extend with LiDAR and vision modalities to broaden applicability across perception–control pipelines.

Abstract

The development of robust learning-based control algorithms for unstable systems requires high-quality, real-world data, yet access to specialized robotic hardware remains a significant barrier for many researchers. This paper introduces a comprehensive dynamics dataset for the Mini Wheelbot, an open-source, quasi-symmetric balancing reaction wheel unicycle. The dataset provides 1 kHz synchronized data encompassing all onboard sensor readings, state estimates, ground-truth poses from a motion capture system, and third-person video logs. To ensure data diversity, we include experiments across multiple hardware instances and surfaces using various control paradigms, including pseudo-random binary excitation, nonlinear model predictive control, and reinforcement learning agents. We include several example applications in dynamics model learning, state estimation, and time-series classification to illustrate common robotics algorithms that can be benchmarked on our dataset.
Paper Structure (4 sections, 5 figures, 2 tables)

This paper contains 4 sections, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. A Dataset for Dynamics Learning
  3. Usage Examples
  4. Conclusion and Outlook

Figures (5)

  • Figure 1: Experimental recording setup for the Mini Wheelbot dataset.
  • Figure 2: Left: The same reference (Yaw Circles) on the three different surfaces contained in the dataset. Right: Python snippet on how to load the dataset.
  • Figure 3: Autoregressive rollout of learned model vs. measured data (MD).
  • Figure 4: Offline evaluation of an estimator against Vicon ground truth data.
  • Figure 5: Classification accuracies over sequence lengths.