A5101626334- Robotic Path Planning Algorithms
- Human Pose and Action Recognition
- Robotics and Sensor-Based Localization
- Advanced Image and Video Retrieval Techniques
- Multimodal Machine Learning Applications
- Low-power high-performance VLSI design
- Hand Gesture Recognition Systems
- Robot Manipulation and Learning
- Formal Methods in Verification
- Tactile and Sensory Interactions
- Semiconductor materials and devices
- Reinforcement Learning in Robotics
- Video Surveillance and Tracking Methods
- AI-based Problem Solving and Planning
- Advancements in PLL and VCO Technologies
- Virtual Reality Applications and Impacts
- Advanced Vision and Imaging
- Neural Networks and Applications
- Stroke Rehabilitation and Recovery
- Balance, Gait, and Falls Prevention
- Human Motion and Animation
- Physical Unclonable Functions (PUFs) and Hardware Security
- Teleoperation and Haptic Systems
- Sepsis Diagnosis and Treatment
- Distributed Control Multi-Agent Systems
Gwangju Institute of Science and Technology
2022-2024
Seoul National University
2017-2024
Samsung (South Korea)
1993-2020
In this paper, we propose a generative model which learns the relationship between language and human action in order to generate sequence given sentence describing behavior. The proposed is adversarial network (GAN), based on (SEQ2SEQ) model. Using network, can synthesize various actions for robot or virtual agent using text encoder recurrent neural (RNN) an decoder RNN. trained from 29,770 pairs of annotations extracted MSR-Video-to-Text (MSR-VTT), large-scale video dataset. We demonstrate...
We present a novel graph-structured memory for visual navigation, called graph (VGM), which consists of unsupervised image representations obtained from navigation history. The proposed VGM is constructed incrementally based on the similarities among observed images, and these are learned an unlabeled dataset. also propose framework that can utilize to tackle problems. By incorporating convolutional network attention mechanism, agent refers navigate environment while simultaneously building...
Trunk rehabilitation exercises such as those for remediating core stability can help improve the seated balance of patients with weakness or loss proprioception caused by diseases stroke, and aid recovery other functions gait. However, there has not yet been any reported method automatically determining parameters that define exercise difficulty on a trunk robot (TRR) based data patient's demographic information, balancing ability, training sequence, etc. We have proposed machine learning...
In this paper, we propose a novel distributional reinforcement learning (RL) method which models the distribution of sum rewards using mixture density network. Recently, it has been shown that modeling randomness return leads to better performance in Atari games and control tasks. Despite success prior work, limitations come from use discrete distribution. First, needs projection step softmax parametrization for distribution, since minimizes KL divergence loss. Secondly, its depends on...
DLL and improved I/O circuits are for 500 Mb/s/pin DDR SDRAM. This digitally-controlled has inherent duty cycle correction capability, enabling fast re-locking upon standby-mode exit. Data input circuits, such as internal delay control digital sense amplifier, reduce setup/hold window to 0.3 ns. The output data driver 62% decreased pattern-dependent skew.
This article introduces a robust and safe path planning algorithm in order to satisfy mission requirements specified linear temporal logic (LTL). When is planned accomplish mission, it possible for robot fail complete the or collide with obstacles due noises disturbances system. Hence, we need find against disturbances. We introduce algorithm, which maximizes probability of success accomplishing given by considering disturbances, while minimizing moving distance robot. The proposed method...
In this paper, we tackle the problem of image-goal navigation which is a crucial robot task but also hard especially when there exist obstacles and limitations on field-of-view (FoV) camera. Conventional visual servoing approaches require depth information camera parameters, are susceptible to FoV loss, while previous learning-based depend unrealistic dense reward function train agent with reinforcement learning. To end, propose novel approach simultaneously utilizes self-supervised local...
This paper presents an UAV control scheme under mission specifications. A is specified in linear temporal logic (LTL) formula. point-to-point local planner designed to search a trajectory satisfying the given specification so that low-level controller of quadrotor can follow trajectory. The adopts Lyapunov functions, and has following properties: (1) It reactively generates with respect current state. (2) generated reflects reference path produced during offline planning procedure. proposed...
We investigate a path planning algorithm for generating robust and safe paths, which satisfy mission requirements specified in linear temporal logic (LTL). When robots are deployed to perform mission, there can be disturbances cause failures or collisions with obstacles. Hence, needs consider safety robustness against possible disturbances. present algorithm, maximizes the probability of success accomplishing given by considering robot dynamics while minimizing moving distance robot. The...
This paper describes a new skew-insensitive U0 scheme for high bandwidth processor-memory communication, which alleviates the interchip skew problem. High speed transmission up to 770Mbaud was obtained with multiphase clocks generated by phase-locked loop circuit. Interchip can be adjusted dual delay-locked based receiver. It is fabricated 0.9pm CMOS process.
In this paper, we tackle the problem of 6-DoF grasp detection which is crucial for robot grasping in cluttered real-world scenes. Unlike existing approaches synthesize data sets and train quality networks with input representations based on point clouds, rather take a novel hierarchical approach does not use any data. We cast as arm approaching direction selection using 4-DoF algorithm, by exploiting fully convolutional network evaluating an direction. To select best highest quality, propose...
A new high cost-performance NAND Flash memory using 3-level MLC and virtual page cell architecture has been proposed. It provides 20% die size saving 3 times fast program speed compared to conventional SLC MLC, respectively. The proposed method can be a good choice of the market demanding both low cost performance as well reliability
Quadrotor unmanned aerial vehicles (UAVs) have seen a surge of use in various applications due to its structural simplicity and high maneuverability. However, conventional control methods using joysticks prohibit novices from getting used maneuvering quadrotors short time. In this paper, we suggest the wearable device, such as smart watch, new remote-controller for quadrotor. The user's command is recognized gestures 9-DoF inertial measurement unit (IMU) device through recurrent neural...
It is presented the first read-out IC (ROIC) for a TFT-based full-screen optical fingerprint (FP) sensor placed under OLED display. The ROIC supports both fast-scan mode to acquire 256×256 pixels FP image in anywhere on display, and full-scan (2,560×1,280 pixels). 2 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">nd</sup> order correlated double sampling (CDS) noise compensation techniques are used low leakage immunity. A low-cost single-slope...
Locomotion gesture classification in virtual reality (VR) is the process of analyzing and identifying specific user movements real world to navigate environments. However, existing methods often necessitate use wearable sensors, which present limitations. To address this, we utilize a high-resolution carpet-type tactile sensor as foot action recognition interface, was previously unexplored context locomotion classification. This interface can capture user's pressure data detail distinguish...
Sepsis is one of the most life-threatening medical conditions. Therefore, many clinical trials have been conducted to identify optimal treatment strategies for sepsis. However, finding reliable remains challenging due limited-scale tests. Here we tried extract sepsis policy from accumulated records. In this study, with our modified deep reinforcement learning algorithm, stably generated a patient artificial intelligence model. As training data, 16,744 distinct admissions in tertiary...
The rapid increase in processor performance and memory density has created the need for high bandwidth DRAM a variety of applications. A 3.3V 2M x8 synchronous is designed with typical data rate 125 Mbyte/sec using internal column address sequencing, pipelined 2 bit prefetch variable output latching scheme. It supports operating frequencies up to MHz.