SNU Robotics Lab서울... · 2020. 9. 29. · SNU Robotics Lab Lab Introduction 2020. Members ......
Transcript of SNU Robotics Lab서울... · 2020. 9. 29. · SNU Robotics Lab Lab Introduction 2020. Members ......
SNU Robotics LabLab Introduction 2020
Members
▪ BS EECS, MIT, 1985▪ PhD Applied Math., Harvard U, 1991▪ Asst. Prof., UC Irvine, 1991-1994▪ Professor, SNU, 1995-present▪ Adjunct Prof., Interactive Computing
Dept., Georgia Tech, 2009-2013▪ HKUST Robotics Institute, 2016-present▪ IEEE Fellow, IEEE RAS Distinguished
Lecturer, SNU Teaching Award 2014▪ Editor-in-Chief, IEEE Trans. Robotics▪ Edx Course Developer (Robot
Mechanics and Control I, II)▪ Co-Author of “Modern Robotics:
Mechanics, Planning, and Control”. 2017
▪ President of IEEE RAS., 2022-2023
Prof. Frank C. Park
13 PhD Candidates, 8 MS Candidates
Research Areas
Robotics
Machine Learning
ComputerVision
Robotics
• Physically consistent multibody inertial parameter identification using geometric algorithm
Robot Modeling
• RRT algorithm that uses an integral function of control effort in a vector field defined in the configuration space
• Adaptive stepsize RRT algorithm which solved cumbersome stepsize-tuning issue
Motion Planning
• A minimum attention control law for ball catching
• Energy based performance criteria for the dynamics-based optimization of robot trajectories
Robot Control
Randomized path planning on vector fields
A Geometric Algorithm for Robust Multibody Inertial Parameter Identification
Dynamics-Based Robot Motion Optimization
Machine Learning
• Coordinate invariant mapping functionals and distortion measures for assessment of closeness of mapping to isometry
• Metric learning for enhanced reliability and less biased estimation
Mathematical foundations of machine learning
• Prediction model for exoskeletons- Enables natural movement that better supports human gait
• Collision detection based on data retrieved from momentum observer
Machine learning in robotics
Industrial applications
Time-series anomaly detection
Original loss
Adversarial loss
Original dataset
Outlier
Negative dataset
Synthetic Outlier
Region represented by dictionary
Movement Prediction
• Time series anomaly detection using adversarial dictionary learning- Learns optimal dictionary that sufficiently expresses time series data- Adversarial training constrains dictionary from including expressions for anomalies
Computer Vision
Object detection• Novel formulations for object detection problem based on
measure theory and information geometry
Vision inspection• Neural network compression via transfer learning for automated
machine vision inspection Object detection on aerial images
Machine vision inspection
Projects
Kinodynamic Model Identification: A Unified Geometric Approach
Introduced a unified method that identify kinematics and dynamics parameter to reduce calibration error
Deep Reinforcement Learning Algorithm Development for Industrial Robot
Unstable motion
Aggressive exploration
Performance
Admittance environment
Motion planning
Velocity control
Acceleration control
Developed safe, efficient reinforcement learning algorithm for position control of industrial robot
Locomotion Characteristics Analysis and Motion Control of Terrestrial Organisms
Mimic the movement of terrestrial organisms and optimize feed back control
Water strider robot simulation Jumping-gliding robot simulationRHex robot simulation
Website
http://robotics.snu.ac.kr/fcp/