As Design Team Lead, a bipedal humanoid robot was developed to perform dynamic motions such as running, balancing, dancing, cartwheels, and backflips. The mechanical system included a low-reduction gearbox, full-body mechanism design, and motor specification selection optimized for performance. In addition, cross-functional collaboration with software and AI teams enabled co-design of the kinematic structure for motion retargeting, including optimization of waist and ankle parallel mechanisms, joint range definition, and actuator placement. Also, design work involved CAD modeling, tolerance analysis, and Finite Element Analysis (FEA), followed by prototype builds, testing, and validation. The system was developed using multi-material 3D printing (including titanium and TPU) and supported by mold design for scalable production, incorporating Design for Manufacturing (DFM) principles.

 We investigated and developed actuator systems with a focus on gearboxes as a core component, alongside motor architectures incorporating novel technologies. Through iterative design, analysis, and experimental validation, custom gearboxes were realized as fully functional prototypes. These included a low-reduction planetary gearbox designed to withstand high-impact loads arising from dynamic locomotion such as running, balancing, and acrobatic motions, as well as a cycloidal gearbox employing a low-friction bearing configuration to achieve high backdrivability.

 In parallel, we explored an axial flux permanent magnet (AFPM) motor architecture featuring a single-rotor, single-stator configuration. This approach enabled a thin-profile design with reduced axial height, improving space efficiency and facilitating integration into space-constrained robotic systems, including humanoid hip joints and compact mobile platforms.

 Wearable sensing sleeve for multi-DOF ankle joint motions using Sim-to-real transfer learning and musculoskeletal simulation was developed by using the capacitive strain sensors ought to its low time-varying property and high linearity. A novel musculoskeletal simulation-based transfer learning approach was proposed to guarantee a fast calibration with small amount of dataset. 

 We proposed the design of a portable double-piston crank microcompressor, which has a simple structure, static mass of 1.5 kg and does not generate any safety hazards. The design requirements in terms of maximum pressure and flow rate were optimized based on wearable robotic applications. The sound intensity level generated by the developed microcompressor was quite low, which can be used for long-term usage, at maximum flow rate. 

2. Insole-type GRF measurement system

 Insole type GRF measurement system with four GRF sensors can measure large vertical GRF and robust to impact, and commercial accelerometer to detect human intention. To keep the user’s motions from being disturbed, GRF and acceleration can be transmitted to PC via Bluetooth. The center of pressure (CoP) from GRF and three axes acceleration (anterior-posterior, medial-lateral and vertical directions) can be measured by the developed device. The system can be applied to the healthcare field and the virtual reality field.