Kyoung-Dae Kim (김경대)

Associate Professor, Electrical Engineering and Computer Science, DGIST
– Office: E3-212 
– Phone: +82.53.785.6325 
– Email: kkim@dgist.ac.kr

 

Education

  • Ph.D., Electrical & Computer Engineering, University of Illinois at Urbana-Champaign 
  • M.S., Computer Science, University of Illinois at Urbana-Champaign 
  • M.S., Mechanical Engineering, Hanyang University
  • B.S., Mechanical Engineering, Hanyang University

Past Positions

  • Assistant Professor, Electrical & Computer Engineering, University of Denver, USA (September 2013 ~ August 2017)
  • Postdoctoral Research Associate, Electrical & Computer Engineering, Texas A&M University, USA (August 2011 ~ August 2013)

Research

Research Statement: 

Develop theories, tools, and software platforms for autonomous, reliable, and cooperative robotic systems based on control, optimization, real-time systems, reinforcement learning, etc. Application areas of interest are autonomous & connected smart transportation system, autonomous driving system, semi-autonomous robotic systems for intuitive and effortless human-machine interface, motion planning of manipulator for collaborative robotics, cooperative multi-agent systems, aerial/ground/underwater robots for search and rescue mission, etc.
 

Autonomous Driving:

  •  MPC based Autonomous Driving for Adaptive Cruise Control and Obstacle Overtaking
    In this work, we propose a model predictive control (MPC) based motion planner that not only ensures safety but also improves driving comfort. The proposed planner generates path tracking and collision-free maneuvers to ensure safety, and improve driving comfort by minimizing acceleration and jerk. Collision-free maneuvers include vehicle following and overtaking.
  • Pure Pursuit with Dynamic Look-ahead Distance and Sideslip Compensation
    To improve the tracking performance, we propose a pure pursuit-based path tracking algorithm that quickly reduces tracking error and considers the sideslip from vehicle dynamics. For fast tracking error reduction, a dynamic look-ahead distance is designed as a function of speed and path curvature by analyzing the tracking error change according to the look-ahead distance.
  •  ARC Drive
    We built an autonomous vehicle, named ARC Drive, as a research platform to do research related to autonomous driving such as motion planning, control, machine learning, etc. The vehicle is equppped with various sensors for perception such as LiDAR, multiple cameras, GPS, and IMU. It is an on-going work to build the software stack for autonomous driving. 

Connected & Autonomous (CA) Traffic System:

  •  Provable System-Wide Safety and Liveness of CA Ground Traffic
    In this work, we propose an approach based on model predictive control (MPC) for the development of provably collision free autonomous ground transportation systems, and present an autonomous intersection management framework. The MPC approach enables a vehicle to generate its own motion locally in time based on an optimization framework, incorporating constraints based on the states of other vehicles in the neighborhood, the speed limit of a road, the maximum values of acceleration and deceleration, etc. Safety and liveness of the traffic are however system-wide properties, not merely neighborhood properties, and the challenge is to augment this distributed optimization with coordination rules that guarantee overall  system-wide safety as well as liveness of the traffic. We design two vehicle-to-vehicle (V2V) coordination rules, along with a vehicle-to-infrastructure rule, and establish the system-wide safety and liveness of the autonomous traffic based on each vehicle’s MPC motion planner, operating in conjunction with an algorithm that orders vehicles according to their runtime properties. 
  • Resilient CA Intersection Crossing Traffic Control Under V2X Communication
    In this work, we present an mixed integer programming (MIP) based approach for safe, efficient, and resilient autonomous intersection traffic control in realistic vehicle-to-everything (V2X) communication environment. The proposed framework produces the fastest discrete-time trajectory for vehicles who want to cross an intersection. Constraints for safety are designed carefully in the optimization problem formulation to prevent potential collisions during intersection crossings. A novel vehicle-to-intersection (V2I) interaction mechanism is designed to handle imperfect communication characteristics such as packet delivery delay and loss.

Tele-Operation:

  •  Adaptive Hybrid Time Series Forecasting for Driving Data Prediction
    In this work, we propose a hybrid time series forecasting model, named as the Adaptive Multivariate Exponential Smoothing – Recurrent Neural Networks (AMES-RNN), which enables accurate prediction for time series data with non-seasonal and additive trend characteristics. To enhance prediction performance, the optimal smoothing parameters of the Exponential Smoothing (ES) model are estimated and updated online. Here, the parameter estimation is performed through a deep learning-based regression model, and a method for training the regression model is presented. An illustrative simulation result shows the remote driving performance difference between with and without the proposed AMES-RNN compensation under various packet delay and loss.
  •  Autonomous Collision Avoidance of Remote Controlled Multi-Rotor UAV
    In this work, we present a potential function based approach, called the Vehicle-centered Potential Function (VPF), for autonomous collision avoidance of a teleoperated multi-rotor UAV. To fuse command from an operator and the repulsive motion from the VPF, we propose an approach which ensures that the UAV can avoid collisions while tracking the operator’s command as closely as possible. A human-in-the-loop simulation system is implemented using the Virtual Robot Experimentation Platform (V-REP) and the collision avoidance performance of the proposed approach is demonstrated through simulations in various cases.

Manipulator:

  •  A Parallelization Algorithm for Real-Time Path Shortening of High-DOFs Manipulator
    The paths generated by sampling-based path planning are generally not smooth and often generate multiple unnecessary robot posture changes in the task space. To mitigate such issues with a planned path from sampling-based path planners, shortcut-based path shortening algorithms are commonly adopted in the field of robot manipulator path planning as a post-processing step. In this work, we analyze shortcut-based algorithms and propose a new approach based on the idea of parallelism for faster path shortening so that it can be more applicable in environments where a path has to be generated as quickly as possible to avoid collisions with other moving objects around the manipulator. Through performance comparisons in simulations, it is shown that the proposed approach can obtain a well-shortened as well as much smooth path compared to the original path faster than conventional shortcut-based algorithms and an optimization-based approach developed for collision-free path generation.
Comparison of end-effector path and computation time .(Blue line is the original path and red line is the shortened path.)

Past Research Topics:

  • Real-time Middleware for Distributed Control System
    A well-designed software framework is important for the rapid implementation of reliable and evolvable networked control applications and to facilitate the proliferation of networked control by enhancing its ease of deployment. In this work, we address the problem of developing such a framework for networked control that is both real-time and extensible. We enhance Etherware, a middleware developed at the University of Illinois, so that it is suitable for time-critical networked control applications. We introduce a notion of quality of service (QoS) for the execution of a component. We also propose a real-time scheduling mechanism so that the execution of components can not only be concurrent but also be prioritized based on  the specified QoS of each execution. We have implemented this framework in Etherware. We illustrate the applicability of this software framework by deploying it for the control of an unstable system, namely, a networked version of an inverted pendulum control system, and verify the performance of the enhanced Etherware. We also exhibit sophisticated runtime functionalities, such as runtime controller upgrade and migration, to demonstrate the flexible and temporally predictable capabilities of the enhanced Etherware.

  • Hybrid System Analysis and Reachable Set Computation 
    An epsilon-reach set of a hybrid automaton is a set of states such that every state in it is within a distance of some reachable state. We propose an algorithm to compute a bounded epsilon-reach set from a given initial state of a class of deterministic linear hybrid automata that satisfy a certain transversality condition. The proposed algorithm is based on time-sampling. It over-approximates the reachable states at each sampled time instant using polyhedra, and subsequently computes an epsilon-reach set for a bounded time interval using these over-approximations, while reducing the sampling period on-the-fly.

  • Dynamic Simulation and Computed Torque Control of Humanoid Robot 
    In this work, we proposes a model called the gravity-compensated inverted pendulum mode (GCIPM) to generate a biped locomotion pattern that is similar to the one generated by the linear inverted pendulum mode, but accommodates the free leg dynamics based upon its predetermined trajectory. The GCIPM includes the effect of the dynamics of the free leg in a simple manner. We also presents a control method for biped robots based upon the computed torque. Simulation results show that the biped robot is more stable with the walking pattern generated by the proposed method combined with the controller than with the one by the inverted pendulum mode.