Human Robot Interaction

Machine Intelligence Learning and its Applications


For human-robot interaction, robots need to be designed for sensing, thinking, and action with MI (Machine Intelligence). Robots with MI developed based on MIL (Machine Intelligence Learning) are allowed to communicate and collaborate with humans. This talk introduces recent research outcomes of the RIT laboratory on MIL for active knowledge acquisition and adaptive knowledge application. The key concept of MIL, the long-term memory is developed as an integrated multi-memory neural model, in which the episodic memory is designed using a Deep DRN (Developmental Resonance Network) neural model, and the semantic memory is built using 3D Scene Graph. 3D Scene Graph is the world first graph-based environment model in which nodes represent objects and edges characterize the relations between pairs of objects. Based on MIL, Mybot, a humanoid robot developed in the RIT Lab., is developed as a unified intelligent robot and applied in various forms including AI Robot with Task Intelligence, Interactive VQA Robot, IoT-based AI Robot, and AI Recommendation System. For IoT-based AI Robot, in particular, Invisible Keyboard is developed as one of the key modules in the IoT environment. Invisible Keyboard is the world best fully imaginary and layout-free keyboard in which humans can type in an eyes-free manner without a calibration step and a predefined region for typing decodes the human typing inputs with a deep learning model. The RIT Lab. also proposes new MIL models applicable to the industry. RRN (Recurrent Reconstructive Network) and Convolutional RRN are the anomaly detection model for temporal and spatiotemporal data, respectively, and applied to the PCB manufacturing process. D3PointNet (dual-level defect detection PointNet) detects defects of the solder paste printer in SMT. Given a point cloud extracted from a defective solder pattern (DSP) image, D3PointNet performs defect detection in two semantic levels: a micro-level and a macro-level. MarsNet (Multi-label classification for images of various sizes network) is a CNN based end-to-end network for multi-label classification with an ability to accept various size inputs and applied to inspect and classify multiple types of defects occurred in PCB screen printer.


Juyoun Park received her B.S. and Ph.D. degrees in Electrical Engineering from KAIST, Daejeon, Republic of Korea, in 2015 and 2019, respectively. She is currently a Post-doctoral Researcher in the Robot Intelligence Technology (RIT) Laboratory, Information & Electronics Research Institute at KAIST working with Prof. Jong-Hwan Kim. Her research interests include but are not limited to the areas of artificial intelligence (AI) for autonomous agents, including robots, in particular, developing novel machine learning or deep learning methods to allow robots to autonomously recognize and understand the environments. It is also one of her research fields to develop a framework for human-robot interaction (HRI). Dr. Park proposed novel classification networks capable of online incremental class learning, OICRN (Online Incremental Classification Resonance Network) for classification, OIHCRN (Online Incremental Hierarchical Classification Resonance Network) for multimedia recommendation system for HRI and CNN-OICRN mainly for face recognition. She applied the proposed networks to real human-robot interaction applications using a humanoid robot, MyBot developed at RIT Lab. Dr. Park also proposed a new deep learning neural network, MarsNet for a defect diagnosis and classification system in smart factories. Dr. Park is working in the domain of machine learning applied to AI framework mainly for HRI.


Juyoun Park, Ph.D., Post-doctoral Researcher
Information & Electronics Research Institute, College of Engineering, KAIST
291 Daehak-ro, Yuseong-gu, Daejeon 34141, Republic of Korea
Tel: +82-42-350-8874, Mobile: +82-10-9726-0726