•Deep learning based image recognition techniques –various object detections and behavioral prediction model
•Deep learning model hardware accelerator design – model reduction and quantization techniques, FPGA and ASIC implementations
•Cloud control center – reinforced learning DQN based distributed computing scheme for route planning, robot fleet management
•Data fusion techniques for image, Lidar, and ultrasonic array radar

The first image based object detection system enhanced with behavioral prediction for people rich environments. A novel edge AI accelerator design for object detection with a computing capability of 4.73TOPS/W. Dynamic feature map update, optimal routing path planning, and DQN distributed computing scheme for cloud control center. The first ultrasonic array radar for DoA and distance estimation.

We provide enabling technologies of autonomous movers to move freely in people rich environments such as shopping mall. These include
•Image based object detector / classifier featuring AI, machine learning techniques and behavior prediction
•High efficiency and low complexity AI edge processor design supporting dynamic reconfiguration
•Social force aware reinforcement learning scheme for navigation control