Zishuo Wang 王梓硕
Email: zishuowang [at] u.nus.edu   Office: COM3-B1-33

profile photo

I am a second-year M.S. student in Computing (AI track) at National University of Singapore, advised by Prof. David Hsu. My research interest lies at the intersection of computer vision and robotics. In particular, I am interested in robotics perception and planning with their applications in challenging real-world environment.

Currently, I'm working on anomaly detection and failure recovery in visual navigation at Adaptive Computing Laboratory. I obtained my B.E. in Automation and B.E. in Artificial Intelligence from Harbin Institute of Technology in 2022, advised by Prof. Huijun Gao.

I am also eager to venture into diverse fields, e.g. utilizing large foundation models to integrate common sense priors into robot, to seek the vast potential of robotics.

CV  /  Github


Research

My current research topics cover 2D & 3D visual understanding, navigation, and autonomous system design. My research vision is to develop intelligent robots capable of perceiving and acting sensibly in dynamic, uncertain real-world environment.

Failure recovery framework for robust navigation with learned controllers

Designed a comprehensive failure recovery framework consisting of perturb and ask for help, that empowers learned controllers to 1) sense anomalous situations, 2) recover themselves, 3) if they cannot recover, fail gracefully. (More in progress)

The video shows some preliminary results in the context of abrupt observation alterations and instances where the pathway is obstructed. The recovery is guided in two ways: 1) we use our anomaly metric to measure recovery success, 2) we backpropagate the anomaly metric then use Grad-CAM to obtain an anomaly localisation heatmap to guide the selection of recovery actions.

Hatch Identification Method for Bulk Carrier

Contributed to the perception module for automatic coal loading at Tianjin Port, demonstrating the practical applicability of the developed approach.

Anti-photographic Detection Algorithm Based on Semi-supervised Learning

Developed a system with good generalizability and accuracy for detecting various photographic equipment, including different styles of phones, cameras, monitors, etc.




Projects

Table Curling Robot

Created a smart robot car which can play curling on table by integrating visual processing, decision making and control.

Biomedical Image Segmentation for Brain Tumour

Implemented a brain tumour segmentation pipeline with modified U-net and designed dice loss.

Intelligent Vision PTZ for Face Tracking and Alerting

Built an 2-Dof Vision PTZ capable of face recognition, tracking and alerting.

Multi-legged Bionic Robot

Built a quadrupedal/hexapedal robot system capable of avoiding obstacles and line following.






Modified version of template from here. Thanks! Last updated: 30 Nov 2023