The participation of assistive robots in current industrial, as well as daily life scenarios, is increasing significantly. To better enable robots to perform the tasks that people are performing, or intend to perform, we want to investigate a deep learning-based vision solution. 

- Deep Learning 

- Python3, Pytorch

- Computer Vision

- C++ and ROS

This work aims to develop low-cost real-time 3D multiple object tracking frameworks based on camera-radar fusion for multiple people tracking. The purpose is to ensure the safety of AGV navigations in industrial autonomous transportation scenarios, most importantly, to avoid collisions with people.

- Basic knowledge of computer vision

- Basic knowledge of deep learning-based tracker model

- Python 3. Pytorch

- C++

Most of the current human activity datasets focus on end-to-end models, i.e., they directly use RGB images as input and ignore the semantic information of human-object interaction. On the one hand, this model is easier to build, but on the other hand, it is also because of the huge amount of work and expense required to annotate the objects. This work focuses on annotating objects in the human behavior dataset to the extent that it allows our object detection models and human activity prediction models to provide valid semantic information.

• Very high motivation
• Computer Vision Fundamentals
• Programming in python
• Basic understanding of deep learning

We use vision solution to locate the target object for the robot and send the desired object back to the operator to complete the whole process of human-robot interaction.

- Panda arm

- Computer vision

- Human-object interaction prediction

-Grasping

Collect a dataset available for assistive robots in a 3D environment. Meet the visual perception system under human-object interaction detection to serve the robot.

- Basics of Realsense 

- Basics of ROS

- Basics of computer vision