Semi-Implicit Data-Driven Predictive Control for Agile Flying and Beyond

Abstract

State-of-the-art model-based control strategies have demonstrated success in enabling dynamic locomotion behaviors such as flying, hopping, and walking in robotic systems. However, the performance of these behaviors in practice remains inadequate, particularly due to the inherent discrepancies between the modeled dynamics and the physical hardware, which inevitably lead to trajectory tracking errors. To mitigate this issue, we propose a semi-implicit control framework that bridges the model-to-real gap by incorporating a data-driven control approach combined with the existing model-based design. We validate the proposed method on PogoX, a custom-designed multi-modal locomotion robot, demonstrating high-precision hopping and flying behaviors in both simulation and real-world experiments. This semi-implicit control paradigm offers a generalizable solution for improving performance across a broad range of robotic platforms and locomotion behaviors.

Yuhao Huang

MS Student

Yuhao joined the lab in the Fall of 2023 as a MS student.

Yicheng Zeng

MS Student

Yicheng joined the lab in the Fall of 2023 as a MS student.

Xiaobin Xiong 熊晓滨

Assistant Professor

Prof. Xiong is a full-stack roboticist who develop rigorous theories for practical applications.