김대겸 교수

Soft robots have been extensively researched due to their flexible, deformable, and adaptive characteristics. However, compared to rigid robots, soft robots have issues in modeling, calibration, and control in that the innate characteristics of the soft materials can cause complex behaviors due to non-linearity and hysteresis. To overcome these limitations, recent studies have applied various approaches based on machine learning. This paper presents existing machine learning techniques in the so

To perceive user intentions for wearable robots, we present a learning-based intention detection methodology using a first-person-view camera.

This study addresses a view-path-planning problem during 3-D scanning of a large-scale structure based on multiview stereo (MVS) for unmanned aerial platforms. Recently, most studies have adopted an explore-then-exploit strategy for 3-D scanning. The strategy first generates a coarse model from a simple overhead scanning and then plans an inspection path to cover the entire surface of the coarse model. However, even though the inspection path may be optimal, it is difficult to guarantee a comple

Soft wearable hand robots with tendon-sheath mechanisms are being actively developed to assist people with lost hand mobility. For these robots, accurately estimating fingertip forces leads to successful object grasping. An approach can utilize information from actuators assuming quasi-static environments. However, non-linearity and hysteresis with regards to the dynamic changes of the tendon-sheath mechanism hinder accurate fingertip force estimation. This paper proposes a learning-based method

Chronic impairment in the paretic ankle following stroke often requires that individuals use compensatory patterns such as asymmetric propulsion to achieve effective walking speeds needed for community engagement. Ankle exosuit assistance can provide ankle biomechanical benefit in the lab, but such environments inherently limit the amount of practice available. Community walking studies without exosuits can provide massed practice and benefit walking speed but are limited in their ability to ass

The practical operation of wearable robots requires intuitive, compact, yet reliable control interfaces. However, current myoelectric interfaces based on surface electromyography (EMG) often fail to achieve these requirements by demanding multiple sensors and exhibiting unreliable performance under limb posture changes. In this study, we show that a myoelectric interface on the musculotendinous junctions (MTJs) of the flexor digitorum superficialis (FDS) enables reliable control of a robotic glo

For stroke survivors, diminished hand functions limit their ability to perform activities of daily living (ADLs). Recently, soft-robotic gloves have assisted stroke survivors in active rehabilitation by facilitating their finger movements based on intentions expressed through biosignals, such as electromyogram and electroencephalogram. In this regard, helping stroke survivors actively train multiple hand postures can improve hand functions required for ADLs. However, detecting intentions regardi

To enhance wearable robots, understanding user intent and environmental perception with novel vision approaches is needed.

Soft wearable robotic gloves based on tendon-sheath mechanism are widely developed for assisting people with a loss of hand mobility. For these robots, knowing the fingertip forces applied to deformable objects is crucial in successfully grasping them without causing excessive deformations. Existing studies presented methods to predict fingertip force applied to rigid objects only using information from the actuation system. However, forces applied to deformable objects are subject to non-linear

This study introduces a deep learning framework for estimating lower-limb joint kinematics using inertial measurement units (IMUs). While deep learning methods avoid sensor drift, extensive calibration, and complex setup procedures, they require substantial data. To meet this demand, we leveraged an open-source dataset to develop and evaluate three training approaches. The first involved training a model exclusively on data from a single user, resulting in high accuracy for that individual only.

Multi-view stereo (MVS) algorithms have been commonly used to model large-scale structures. When processing MVS, image acquisition is an important issue because its reconstruction quality depends heavily on the acquired images. Recently, an explore-then-exploit strategy has been used to acquire images for MVS. This method first constructs a coarse model by exploring an entire scene using a pre-allocated camera trajectory. Then, it rescans the unreconstructed regions from the coarse model. Howeve

Inertial measurement units offer a cost-effective, portable alternative to lab-based motion capture systems. However, measuring joint angles and movement trajectories with inertial measurement units is challenging due to signal drift errors caused by biases and noise, which are amplified by numerical integration. Existing approaches use anatomical constraints to reduce drift but require body parameter measurements. Learning-based approaches show promise but often lack accuracy for broad applicat