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

Wheelbot Dataset

A large, high-quality dynamics dataset of the [Mini Wheelbot](https://github.com/wheelbot/Mini-Wheelbot). The dataset contains 1 kHz data of all onboard sensor readings, the estimated state, ground-truth pose measurements from a motion capture system, and a third-person view video of the experiment. We perform a variety of experiments using pseudo-random binary excitation signals (PRBS) as setpoints of a linear controller, an MPC for driving, and an RL policy that races along a track. Experiments are performed across multiple hardware instances and on different surfaces. With this dataset, we hope to encourage researchers to use the Mini Wheelbot to benchmark their learning-based control methods, even without access to the real hardware. We include two example implementations of how to use the dataset, i.e., for dynamics learning and state estimation. A detailled description of this dataset is available in the brief on arxiv.