Jiawei Fu

Rämistrasse 101

8092 Zürich

I am currently a master’s student in Robotics, Systems and Control advised by Prof. Marco Hutter at ETH Zurich. My research goal is to automate mobile robot navigation with visual inputs. Specifically, I am interested in

learning useful representations to encode task-related objects
increasing the robustness of neural networks against unseen objects and unpredictable system errors
incorporating safety constraints into reinforcement learning algorithms

I am now conducting my master’s thesis under the supervision of Prof. Amir Zamir at Visual Intelligence and Learning Lab, EPFL. Before that, I undertook a research internship supervised by Prof. Dengxin Dai at Vision for Autonomous Systems, Max Planck Institute for Informatics. I also did a student internship advised by Prof. Davide Scaramuzza at Robotics and Perception Group, University of Zurich. Besides, I finished my semester project under the supervision of Prof. Marco Hutter at Robotic Systems Lab, ETH Zurich.

Prior to my master’s study, I received my bachelor’s degree in Mechanical Engineering (Elite Program) at Tsinghua University. During my undergraduate study, I received various awards and scholarships, such as National Scholarship (top 2%), Comprehensive Excellence Award (top 10%), and so on.

For more information about my experience, please refer to my CV page or my one-page CV (PDF).

News

Dec 1, 2022	I will join Visual Intelligence and Learning Lab for my master’s thesis.
Apr 1, 2022	I will join Vision for Autonomous Systems as a research intern.
Sep 1, 2021	I will join Robotics and Perception Group as a student intern.

Selected publications

Learning Deep Sensorimotor Policies for Vision-based Autonomous Drone Racing

Jiawei Fu, Yunlong Song, Yan Wu, Fisher Yu, and Davide Scaramuzza

arXiv preprint arXiv:2210.14985, 2022

Abs arXiv Bib PDF Video

Autonomous drones can operate in remote and unstructured environments, enabling various real-world applications. However, the lack of effective vision-based algorithms has been a stumbling block to achieving this goal. Existing systems often require hand-engineered components for state estimation, planning, and control. Such a sequential design involves laborious tuning, human heuristics, and compounding delays and errors. This paper tackles the vision-based autonomous-drone-racing problem by learning deep sensorimotor policies. We use contrastive learning to extract robust feature representations from the input images and leverage a two-stage learning-by-cheating framework for training a neural network policy. The resulting policy directly infers control commands with feature representations learned from raw images, forgoing the need for globally-consistent state estimation, trajectory planning, and handcrafted control design. Our experimental results indicate that our vision-based policy can achieve the same level of racing performance as the state-based policy while being robust against different visual disturbances and distractors. We believe this work serves as a stepping-stone toward developing intelligent vision-based autonomous systems that control the drone purely from image inputs, like human pilots.
@article{fu2022learning, title = {Learning Deep Sensorimotor Policies for Vision-based Autonomous Drone Racing}, author = {Fu, Jiawei and Song, Yunlong and Wu, Yan and Yu, Fisher and Scaramuzza, Davide}, journal = {arXiv preprint arXiv:2210.14985}, year = {2022}, }