Hyun-Min Joe

This paper proposes a reinforcement learning (RL)-based control method for robust path tracking of a 4-wheel independent steering (4WIS) mobile robot in outdoor rough terrain environments. Traditional wheeled robots typically suffer from limitations including mobility constraints in narrow spaces, path deviations caused by ground slip, and reduced traction on rough terrain. To address these challenges, we designed a 4WIS mobile robot and implemented an architecture that independently controls the steering and driving of each wheel. The RL state space is defined by look-ahead path information, robot pose, velocity, and tracking errors, while the action space consists of target angular velocity and steering angle. To ensure robust performance, we applied random path and terrain generation and implemented domain randomization for sensors and actuators based on empirical GPS and motor data. The proposed controller was validated against the Pure Pursuit algorithm through dynamic simulations and real-world experiments. In simulations mimicking outdoor terrain, the controller reduced lateral and heading RMSE by 6.32% and 16.00%, respectively. In actual outdoor environments, it reduced these errors by 21.54% and 4.78%, respectively. These results demonstrate that the proposed controller provides superior robust tracking performance in unstructured outdoor environments.

This paper proposes a learning-based contact force controller using deep neural networks (DNN) and a PI controller. Stable contact force control between the foot and the ground is essential for humanoid robots to maintain balance during bipedal walking. While admittance controllers have been extensively employed for contact force control in humanoid robots, their performance is limited by the high nonlinearity inherent in robot systems. To overcome these limitations, we propose a deep neural network (DNN)–based inverse model, which leverages input–output data that inherently capture system nonlinearities. The proposed learning-based contact force controller computes the target foot height based on the target force, measured force, and measured foot height, without relying on a dynamic model of the articulated robotic leg. Furthermore, a PI controller is integrated to mitigate steady-state errors. Experimental comparisons between the proposed controller and an admittance controller were conducted using an articulated robotic leg. Compared with an admittance controller, the proposed method reduced overshoot by 96% and settling time by 61% on average in step responses and decreased force-tracking RMSE by 66.3% on average across both step and sinusoidal experiments.

In this letter, we propose learning-based force control of twisted string actuators (TSAs) using a neural network-based inverse model. A learning-based force controller is designed using the input and output data of TSAs without a dynamic model of TSAs. Furthermore, the neural network-based inverse model is utilized to reduce model errors and handle nonlinearities between the inputs and outputs of the TSAs. The trained neural network-based inverse model is directly implemented as a force controller for the TSAs. Additionally, we propose data collection methods utilizing three types of inputs to improve the performance of the proposed controller. To verify the improved performance resulting from the proposed data collection methods, we compared the performance of the learning-based force controller for each dataset in the TSA hardware. We then selected the dataset with the best performance among the proposed inputs through experiments. Additionally, to verify the performance of the learning-based force controller, a reference force tracking experiment was performed and compared with a proportional-integral-derivative (PID) controller and feedback linearization. The learning-based force controller utilizing the selected input-based dataset demonstrated higher force tracking performance than the other controllers. Consequently, the TSA's learning-based force control demonstrates robust force control performance.

An upper extremity device can help humans when they are walking with a load. However, when using such a device that does not have lower limb parts, the upper and lower bodies of the wearer should support the weight of the load. To solve this limitation, this study proposes an unpowered wearable exoskeleton that can reduce the weights of loads felt by the wearer when they are walking. Using supporting force, trunk extension, and loaded walking tests, for a load weight of 12 kgf, the exoskeleton could reduce the weight transmitted to the wearer by 7 kgf, decrease the human trunk extension angle by up to 1.8° compared to that in case of an upper extremity device, and lower the net metabolic rate of the wearer by up to 47.41% during loaded walking. Especially, such reductions in the net metabolic rate have not been reported to date. These three primary results indicate that the exoskeleton in this article can be used to augment the human load-carrying capability and solve the structural problems associated with the use of an upper extremity device. Furthermore, because the exoskeleton requires no power sources and actuators, it is energy-efficient and can be used in the water environments. This exoskeleton is economical owing to its simple structure. Moreover, the model verified in this paper can be utilized in other similar systems having a spring mechanism to optimize the systems.

This paper proposes a device that can change the payload of an unpowered lower-limb exoskeleton supporting the weights of humans and loads. Our previous exoskeletons used a cam–follower structure with a spring applied to the hip joint. This exoskeleton showed satisfying performance within the payload; however, the performance decreased when the payload was exceeded. Therefore, a payload adjustment device that can adjust the wearer’s required torque by easily applying it to the cam–follower structure was developed. An exoskeleton dynamic equation that can calculate a person’s required joint torque given the required payload and the wearer’s posture was derived. This dynamic equation provides a guideline for designing a device that can adjust the allowable joint torque range of an unpowered exoskeleton. In the Adams simulation environment, the payload adjustment device is applied to the cam–follower structure to show that the payload of the exoskeleton can be changed. User convenience and mass production were taken into account in the design of this device. This payload adjustment device should flexibly change the payload of the level desired by the wearer because it can quickly change the payload of the exoskeleton.

To achieve human-level object manipulation capability, a robot must be able to handle objects not only with prehensile manipulation, such as pick-and-place, but also with nonprehensile manipulation. To study nonprehensile manipulation, we studied robotic batting, a primitive form of nonprehensile manipulation. Batting is a challenging research area because it requires sophisticated and fast manipulation of moving objects and requires considerable improvement. In this paper, we designed a batting system for dynamic manipulation of a moving ball and proposed several algorithms to improve the task performance of batting. To improve the recognition accuracy of the ball, we proposed a circle-fitting method that complements color segmentation. This method enabled robust ball recognition against illumination. To accurately estimate the trajectory of the recognized ball, weighted least-squares regression considering the accuracy according to the distance of a stereo vision sensor was used for trajectory estimation, which enabled more accurate and faster trajectory estimation of the ball. Further, we analyzed the factors influencing the success rate of ball direction control and applied a constant posture control method to improve the success rate. Through the proposed methods, the ball direction control performance is improved.

This study proposes a biped-robot pelvis-kinematics estimator based on the touch-point updating method. Because the pelvis frame is used as the base coordinate for the control, the kinematics of it with respect to the global frame should be precisely estimated. To this end, it was necessary to know where the robot made contact with the ground. The touch-point concept was introduced as the temporal contact-point, which was instantly the robot’s rotation center. By updating this point, the biped's global pelvis-kinematics could be estimated. The proposed estimator was implemented into the actual robot, and its superiority was verified through ground-truth data.