Pervasive Artificial Intelligence Research (PAIR) Labs

Robotics & Sensing Technology

Pervasive Artificial Intelligence Research (PAIR) Labs

Robotics & Sensing Technology

NCTU – Professor Jiun-In Guo

Principal Investigator: Professor Jiun-In Guo

Research：

Embedded AI Deep Learning Technology for ADAS/Niche Self-Driving Applications

Core Technologies

This project goes on developing embedded AI deep learning technology focus on the ADAS/Self-driving applications. We develop the technology from five aspects, including automatic object labeling tool and labeled datasets, deep learning software and hardware technology development, various ADAS/Self-driving object detection and behavior predition technology, self-driving control technology as well as virtual simulation environment establishment for ADAS/Self-driving applications.

Applications

ADAS system, self-driving vehicles, self-driving wheelchair, industrial AGV, self-driving farming machines

Technology Transfer

1. Embedded MTSAN model for deep learning object detection/segmentation in ADAS applications
2. Embedded camera/radar sensor fusion technology
3. ezLabel, a fast deep learning object labeling tool
4. ezQUANT, a bit-accurate deep learning model quantization/training tool for deep learning accelerators
5. ezHybrid-M, a hybrid fixed point/binary CNN model training tool for low-power object detection/classification applications
6. Modularized self-driving hardware/software systems for niche self-driving applications

Technologies with Potential for Technology Transfer or Industry-Academia Collaboration

1. A vector-map free self-driving system based on Autoware
2. Embedded object detection/tracking for ADAS/Self-driving
3. Embedded instance/semantic segmentation

Proposal for Industry-Academia Collaboration

1. Deep learning ADAS system
2. Self-driving hardware/software technology for niche market applications
3. Camera/radar sensor fusion system for smart transportation
4. Embedded AI deep learning technology development
5. Automatic labeling tool for object detection/behavior analysis in camera/radar/lidar domains
6. Hybrid fixed point/binary CNN hardware accelerator for edge AI applications
7. 360 video SLAM and camera/radar fusion SLAM for self-driving applications
8. Embedded moving object behavior analysis