UW Data Science Seminar: Hanquan “John” Wang

“DirectHop: Data-Driven Control and Predictable Locomotion in a 1.0g Direct-Drive Jumping Robot” Abstract: Insect-scale jumping robots have traditionally relied on “store-and-release” spring mechanisms to overcome power density limits. While effective for raw power, these systems are inherently all-or-none and lack the precision required for complex navigation. This talk introduces DirectHop, a 1.0g robot that utilizes a direct-drive architecture to achieve full closed-loop authority and variable jump heights. We will focus on the data science and control workflows that enable this performance at the sub-gram scale. Key technical highlights include the implementation of a 1 kHz control loop and an Exponential Moving Average (EMA) filter (α=0.1) to handle high-frequency switching noise from the motor driver. I will detail our parallel feedforward-feedback control algorithm, which utilizes Ohm’s Law to compensate for terminal resistance, achieving a 20ms rise time critical for millisecond-scale launches. Experimental results demonstrate a highly predictable linear relationship (R^2 ≈ 0.982) between commanded coil current and jump height, a level of controllability previously unseen in microrobotics. We will also examine the validation of our active self-righting system, which maintains a 90% success rate through dynamic Center of Mass (CoM) modulation. Finally, we will discuss modeling for autonomous operation.