Robot Team's Research Area

Transparent Actuation System Control

Non-disabled patients or patients who are capable of voluntary movement might feel discomfort due to the resistance of the actuators. By precisely controlling the interaction force between the robot and the wearer, the robot can ensure natural movement or assist insufficient muscular strength. Designing actuators or control algorithms that allow the precise control of forces between the robot and the wearer is needed. Also, there exists irresistible mechanical impedances and non-linearities in controlling wearable robot which make patients feel uncomfortable. These factors can be removed by suitable compensators. Furthermore, it is necessary to take into account the control performance of the actuator when designing actuators.

Ergonomic Robot Design

Most of the existing exoskeletons do not move like human motion due to the approach from the robotics point of view. The robot team is researching the ‘wearable’ exoskeletons starting from the human robot interaction (HRI) perspective. From this point of view, the safety of the musculoskeletal system is an uncompromisable factor. To reduce the risks caused by joint misalignment, bio-inspired knee joint and multi-degree-of-freedom systems have already been researched in Exo-Lab. Furthermore, the robot team is working on the ergonomic design factors such as weight, size, and kinematic structure of exoskeletons to enhance wearability. Also, to address the muscle weakness issue for stroke patients and the elderly, assistance is strongly recommended. However, the wearable robot system itself could degrade the functionality of the muscle due to the robot’s weight. The robot team is researching cable-driven robots too, in that the weight could be distributed to other parts from the affected side so that the weight on the patients could be decreased.

Actuator Module Design

The actuator, which generates torque and transfers power, is a core mechanical element of every robot. Moreover, exoskeletons require actuators with small, lightweight, and high-power density since it directly interacts with human. Robot team is developing optimized motor for use in exoskeleton. To meet the requirements, we are researching and developing an Axial Flux Permanent Magnet (AFPM) Motor with high power and torque density per unit volume. The robot team is researching on increasing the torque of motor and to reduce the cogging torque to improve the control performance. While driving this motor, the specific driver is required to convert the single-phase battery voltage to three-phase and apply the appropriate commutation for each step. A relatively high driver output is required to assist walking and behavior, so the motor driver must switch high currents at a stable and fast speed. A sufficiently small form factor is also essential for the application to wearable robots.

Human-Oriented Sensor Design

Recognition of human gestures plays an important role in wearable robots. Electromyography(EMG) is one of the most common method but it is too sensitive to environmental disturbances which makes it difficult to use for wearable robots. Robot team is developing a new method for recognizing the muscular activities based on air pressure and air-bladders. The interaction force between robot and environment is also necessary for precise control and the performance verification of wearable robots. In particular, multi-axis ground reaction force has to be measured for gait analysis and to measure this, an air pressure-based soft force measurement unit is being studied and developed in robot team.

AI Team's Research Area

Human Motion Research

Although reinforcement learning has accelerated the development of humanoid robots, it has been difficult to use reinforcement learning for wearable robots until now. Since a wearable robot, unlike a humanoid, has irresistible interaction forces with a human, it is necessary to model a human as well as the wearable robot on a simulator. Then, it should be possible to simulate how humans react to the external forces and obstacles by implementing human motion control mechanisms rather than modeling only their appearance. For this, the Human Motion Control Theory(HMCT) will be studied. Each element necessary for human movement is divided into C(cerebrum motor cortex), G(skeleton), M(muscle), S(spinal cord cortex), and H (proprioceptor), and each will be modeled as an aspect of control engineering.

Human robot integrated simulation

After finding out the human motion control mechanisms through human motion research, in order to apply reinforcement learning to wearable robots, humans and robots should be implemented on a simulator. While implementing soft tissues such as human muscles and tendons, we will build an environment that can precisely implement human-robotinteraction. We plan to use Nvidia's Isaac sim or deepmind's mujoco.

Torque and position optimization

Despite the various impairment of the patient, the control strategy should be optimized individually to fully overcome the gait disorder. For that, in our AI team, reinforcement learning-based optimization will be performed even if the optimizable trajectory is a torque for the rehabilitation or human augmentation or position for the paraplegic assistive robot. The AI team will develop the network-based method which can smoothly implement the intention and mimic the human’s natural dynamic movement, avoiding obstacles, and also the human body’s reaction motion including balance and collision. Furthermore, the total reward function will be carefully modified by bringing the medical/biomechanical knowledge to ensure the musculoskeletal system's safety also.