Weixian Shi
Weixian Shi
Phone: +49 15257477188 Email: weixian_shi@outlook.com
Education
2019 - 2022 MSc, University of Bonn Computer Science, Germany, English
2014 - 2018 BSc, Shandong University Software Engineering, China, Bilingual
Work
2021 - 2022 Humanoid Robots Lab Student research assistant
2020 - 2022 Fraunhofer IAIS Software engineer
2017 - 2017 JD.com Java developer intern
Research Projects
Re-implementation of a Job Schedule paper based on Deep Reinforcement Learning
- Paper: “Research on Adaptive Job Shop Scheduling Problems Based on Dueling Double DQN”
- Four days from understanding the field to reproducing, to achieve basically the same effect as the paper (sometimes even better)
- Link: https://github.com/weixians/JobSchedule
Subgoal-driven robot navigation based on Deep Reinforcement Learning
-
Master thesis project, with project link available.
- A hierarchical navigation framework for a lidar equipped robot navigating in indoor environments with unknown static obstacles and pedestrians.
- The high-level agent predicts subgoal position for next fixed time period, while the low-level agent aims to achieve the subgoal.
- The result will be rearranged and submitted to ICRA 2023.
-
Techniques: Deep reinforcement learning (DDPG, TD3), Pybullet, ROS, Python, C++
Enhanced Spatial Attention Graph for Motion Planning in Crowded, Partially Observable Environments
-
Paper accepted by ICRA 2022, with link available.
- Propose a new neural network based on reinforcement learning for human-aware partially observable environments.
- This approach outperforms the previous state-of-the-art methods.
-
Techniques: Deep reinforcement learning (Double DQN, imitation learning), Python
Software Development Work projects
Investment relation map for JD stock app
- Company: JD.com
- A Spring project providing data for JD stock app.
- Integrate data from multiple MySQL databases, find and aggregate the relationship between major shareholders and companies in massive data,
and store the results in Redis.
- Techniques: Quickly and efficiently aggregate useful information from large amounts of data.
Web match system for electronic laser shooting sports events
- Company: Zhuhai Guangge Technology
- A web-based match system to hold sports matches, collect scores from Android app and display the real-time ranking information.
- Separating the system into several modules, in which Spring cloud is used to help communicate with each other.
- Websocket is used to maintain the messaging between frontend, Java backend and Android app.
- Techniques: Spring boot, Vue.js, Websocket, Mysql, Redis, MongoDB
Skill Summary
- Deep Reinforcement Learning, Machine Learning, Deep Learning
- Familiar with Python, Java, Android, Git, Linux, Pytorch, Redis
- Experienced on C++, Vue.js, ROS, Tensorflow, Mysql, MongoDB
- Human Languages: Chinese (native speaker), English (fluent)
- High passion on learning new technologies
- Good team player
Publications
Shi, Weixian, et al. “Enhanced Spatial Attention Graph for Motion Planning in Crowded, Partially Observable Environments.” 2022 International Conference on Robotics and Automation (ICRA). IEEE, 2022.
© 2022 weixians.github.io