Height Control and Optimal Torque Planning for Jumping With Wheeled-Bipedal Robots

A team of Chinese researchers (Zhuang et al.) has tackled a long-standing challenge in legged robotics: accurate height control during jumping for wheeled-bipedal platforms. These robots combine wheels for efficient locomotion with legs for agility, but jumping introduces underactuation, nonlinear dynamics, and instantaneous ground impact. Existing controllers often over-jump to guarantee safety, causing excessive motor wear, higher impact forces, and wasted energy. The team first developed a wheeled-bipedal jumping dynamical model (W-JBD) that provides a theoretical basis for height control, but its torque profile had sharp discontinuities unsuitable for real hardware.

To solve that, the researchers introduced Bayesian optimization for torque planning (BOTP). Instead of requiring an exact dynamic model, BOTP treats torque planning as a black-box optimization problem and finds the optimal continuous torque curve in just 40 iterations on average (using the W-JBD model to bound the search space). In Webots simulations, BOTP achieved 82.3% lower height error and 26.9% lower energy cost compared to baseline methods. The continuous torque profile also means less mechanical stress. This work, accepted at ICARM 2021, paves the way for deploying efficient, precise jumping on real wheeled-bipedal robots, with potential applications in search-and-rescue, inspection, and delivery.