MRI medical robots assist patients in MRI environments, enabling doctors to make therapy plans based on MRI images. However, traditional electric actuators with magnetic motors are unsafe due to the strong magnetic field. Non-magnetic actuators, like hydraulic and ultrasonic motors, are non-backdrivable, risking patient safety. To address this, we developed an MRI-compatible series elastic actuator using an ultrasonic motor and a non-metal spring. This design ensures compliance and precise force control, enhancing safety and comfort during patient interaction. Compared to velocity-controlled robots, these force-controlled robots offer better real-time responsiveness to operator commands and patient actions.

Human strength augmentation exoskeletons amplify human-applied forces, making it easier to carry heavy loads. Control is challenging due to human biomechanics' uncertainty. Our research improved accuracy by discovering human joint complex stiffnesses. We developed a fractional-order exoskeleton controller to handle these stiffnesses, achieving a tenfold strength increase while maintaining stability, unlike previous linear controllers, which were unstable at higher factors. My controller is a one-parameter system, allowing customization for different individuals. Additionally, we developed an online learning method to tune the exoskeleton controllers in real time, providing efficient augmentation continuously.

This research project investigated human biomechanics and robot control that mimics it. Previous studies showed humans adapt to changes in external load inertia, maintaining stability, whereas robots can become unstable. The reason for human adaptivity to uncertain inertia was unknown. Through human subject studies, we discovered hysteretic damping in joint dynamics and developed a model incorporating this nonlinear damping, replacing the traditional linear model. This model showed human motion stability is unaffected by external load inertia. To replicate this in robots, we created a switching linear controller that approximates the noncausal human hysteretic damping, enabling robots to adapt to uncertain environments.

Human-robot teaming enables joint control of a system by both a human and a robot, providing effective solutions for various applications. This collaboration is challenging because the robot must plan its trajectory based on non-convex safety constraints defined by the human’s objectives, which are not directly provided. We addressed these safety issues with a two-layer control framework. The upper layer uses a sampling-based method to decompose non-convex problems into convex sub-problems, while the lower layer employs a linear matrix inequality optimization algorithm for feedback control. Unlike previous methods, our approach ensures safety during trajectory transitions and infers human objectives using a Bayesian method, allowing the robot to assist and correct unsafe human actions.

In this research project, we developed robot control algorithms that ensure safe operations by evaluating a barrier function, which indicates proximity to safety bounds. Finding an effective barrier function and a corresponding controller is challenging. We created a linear matrix inequality (LMI) synthesis method to simultaneously generate a valid barrier function and a safety-enforcing controller, unlike previous studies that addressed them separately. Our method guarantees the existence of a safety-enforcing controller once the barrier function is identified. Additionally, we extended my LMI synthesis method to estimate the barrier function using partial state information, allowing safety assessment and correction in human-in-the-loop robotic systems.