Using dynamic simulations, precise design requirements for key joints, including the hip and knee, were established to ensure that the robot could meet the torque and power needs of walking. Based on these results, high-torque, high-power density motors were implemented, and custom planetary gear systems for the hip and knee joints were developed. These systems achieved the required torque output while minimizing friction and nonlinearities.

During the development process, robust actuation modules for key joints such as the hip and knee were designed. Dynamic simulations were used to determine precise design requirements, and high-torque, high-power density motors were implemented to meet these needs. Additionally, custom planetary gear systems were developed to achieve the required torque output while minimizing friction and nonlinearities, ensuring both durability and precise force control. For the ankle joint, a parallel drive module with linear actuators was adopted, effectively balancing high reduction ratios and precise control while accounting for the unique workspace constraints of human ankle movements.

Research efforts have been dedicated to advancing the design and validation of wearable robots, focusing on the development of innovative systems that integrate human motor control theory. The work has centered on creating wearable robots tailored for individuals with mobility impairments, particularly those with complete lower-body paralysis, enabling them to perform daily activities independently. To achieve this, a 12-degree-of-freedom wearable robot was designed and developed, capable of supporting balanced, autonomous movement without the need for external aids.

Additionally, specialized modules for hip abduction/adduction and rotation were developed to enable smooth directional changes and posture adjustments during walking. The hip rotator module, in particular, was engineered to withstand significant loads and deliver rapid directional shifts, ensuring both mechanical robustness and control accuracy. Each component underwent rigorous validation, including system identification and frequency response analysis, to ensure performance under real-world conditions. Through these efforts, highly reliable actuation systems were achieved, capable of accurately interpreting and replicating human motion.

Beyond hardware development, findings were integrated into comprehensive modeling and simulation environments to validate human-robot interactions. By combining experimental data with precise simulations, theoretical advancements were bridged with practical applications, ensuring the systems' real-world reliability and functionality.

Additionally, efforts have been focused on designing wearable robots that enable individuals with mobility impairments to don the system independently, even while seated in a wheelchair. Through advancements in the prototype wearable robot and docking suit, a novel docking mechanism was developed to ensure precise alignment and stability, allowing seamless donning without external assistance. By combining these practical solutions with precise simulations, theoretical advancements were effectively translated into real-world applications, ensuring that the wearable robot remains both reliable and functionalfor everyday use.

Actuator control performance and robustness are highly sensitive to operating temperature, as variations in temperature affect critical properties such as torque constant, friction characteristics, and electrical parameters. To address this, actuator homeostasis is defined as the ability to maintain consistent output for identical input across a wide range of operating temperatures. One challenge to achieving homeostasis arises from temperature induced variations in the torque constant, which are influenced by changes in magnetic flux density. Traditional compensation methods often rely on static temperature measurements of components, which fail to account for dynamic temperature changes in rotating components such as magnets. To overcome these limitations, a novel thermal model was proposed to estimate unmeasurable magnet temperatures in real-time, and compensation for the resulting torque constant variations was implemented.

A lumped parameter thermal network (LPTN) was proposed to model the thermal dynamics of the actuator, considering key nodes including the stator, rotor, front housing, side housing, and gear. To address unmeasurable components, a temperature observer was incorporated, leveraging measurable node temperatures for real-time estimation. This approach enhances the accuracy of thermal predictions, ensuring reliable actuator performance under varying thermal conditions.

The proposed torque constant variation compensator addresses changes in torque constant caused by temperature fluctuations in the rotor’s permanent magnets. A temperature observer is employed to estimate the rotor’s temperature in real-time, which directly affects the torque constant. By adjusting the reference current based on the estimated temperature, the compensator ensures accurate torque output, even under varying thermal conditions, maintaining actuator performance and stability.

A robust three-dimensional posture estimation system for wearable robots has been developed to enhance the recognition of user intentions, state estimation, and control in human-robot integrated systems. A core aspect of this work involved designing a novel algorithm capable of maintaining high estimation accuracy under dynamic motion and external magnetic disturbances, which are inherent challenges in wearable robot operation.

Traditional IMU-based posture estimation methods often fail under rapid acceleration or impact, as experienced during dynamic movements, and are further compromised by magnetic distortion from the robot's metal frame and motor operation. To address these limitations, a time-varying complementary filter (TVCF) algorithm was designed to combine accelerometer and gyroscope data while dynamically adjusting sensor reliability based on motion intensity. This approach ensured robust posture estimation regardless of the motion state.

The algorithm incorporates a unique filtering structure that operates on quaternion representations. By mapping quaternions to a linear space, applying low-pass and high-pass filters, and converting back to the nonlinear manifold, precise estimation was achieved even in nonlinear dynamic conditions. Additionally, the algorithm dynamically adjusts sensor filtering frequencies in real-time based on motion intensity, favoring accelerometers during low movement and gyroscopes during rapid motion. A flushing mechanism was also implemented to maintain stability during abrupt changes in motion states, preventing filter divergence.

This system was validated by integrating it into a wearable robot’s torso IMU to estimate roll, pitch, and yaw relative to the ground. Using forward kinematics as ground truth, the TVCF demonstrated high accuracy across various dynamic conditions, including walking and trotting. Comparative experiments with commercial IMUs further confirmed the superiority of this approach, especially in mitigating high-frequency vibrations caused by ground impacts during walking.

Research has been conducted on the comprehensive integration and validation of wearable robots designed to restore mobility and independence for individuals with severe physical impairments. A key aspect of this work involved developing a human-robot integration framework that accurately models the interactions between the robot and its user. By incorporating a virtual spring-damper model, the elasticity of wearable components was effectively represented, enabling seamless synchronization between human movements and robotic support. This foundation not only ensured a stable integration but also opened pathways for future research incorporating muscle and neural system models.

Recognizing the unique challenges of wearable robots, a simulation architecture was designed to allow for the independent modeling and control of human and robotic systems. Unlike traditional robotic systems, wearable robots must accommodate the natural, autonomous movements of the user. To address this, a 12-degree-of freedom robotic model with dynamic properties and torque limitations tailored to real-world conditions was created. This was complemented by a detailed human model accounting for various levels of physical impairment, including full lower-body paralysis.

To refine and validate control algorithms, reinforcement learning methods, including adversarial imitation learning, were employed. These methods ensured that the robot replicated human gait patterns intuitively while maintaining user safety. Particular focus was placed on sagittal plane joints, such as the hip and knee, to balance stability, flexibility, and power efficiency. Progressive learning techniques were applied to improve convergence and adaptability under complex conditions, such as uneven terrain and dynamic upper-body movement.

Furthermore, a robust framework for generating and implementing motion trajectories based on model predictive control was developed. Starting with single-step trajectory generation, the framework was expanded to support continuous, smooth gait cycles. This system was optimized for energy efficiency and natural motion, ensuring long term usability and comfort for the wearer. Simulated trajectories were systematically transitioned to real-world applications through rigorous hardware testing, bridging the gap between virtual environments and practical use.

The culmination of this work was the successful demonstration of a wearable robot that enabled an individual with complete lower-body paralysis to walk independently without crutches. This achievement highlighted the effectiveness of the integrated approach to design, modeling, and control. By combining advanced theoretical frameworks with practical validation, this research has set a new standard for the development of high degree-of-freedom wearable robots, offering significant potential to transform lives through enhanced mobility and independence.