Project 69：Deep learning-based Motion Planning in High Dimensional Spaces

Contact Information

Assi. Prof. Xiaoming Duan      

Email: xduan@sjtu.edu.cn

Project Description and Objectives

Highly efficient motion planning is key enabler for robots to be deployed in real-world scenarios. Traditional sampling-based motion planning algorithms often suffer from high time complexity and are not applicable in high dimensional planning spaces with strict real-time performance requirement. Recent advances in deep learning techniques are promising in helping address the complexity of sampling-based motion planning algorithms. The objective of this project is to develop deep learning-based motion planning algorithms that are efficient in high dimensional planning spaces. The developed algorithm will be tested against state-of-the-art learning-based motion planning algorithms on established simulation environments.

Eligibility Requirements

Interested students should be proficient in python and C++ programming and have basic knowledge about deep learning algorithms. Proficiency in English writing and speaking is also mandatory.

Main Tasks

Build motion planning simulation platform for simulating the robots and typical planning environments and implementing motion planning algorithms

Investigate existing learning-based motion planning algorithms and implement them

Develop new learning-based motion planning algorithms and compare them with SOTA

Website

Lab: https://iwin.sjtu.edu.cn/En

School: http://www.seiee.sjtu.edu.cn/