Krispin Wandel
PhD candidate in Control Science and Engineering at SJTU
Computer Vision + Robotic Learning | CVPR | MSc ETH Zurich | Founder of Discuna (adopted at ETHZ)
I am a PhD candidate at the Intelligent Robotics and Machine Vision Lab at Shanghai Jiao Tong University, supervised by Prof. Hesheng Wang. My research focuses on building semantic and physical shape priors from images and videos and leveraging them to improve data-efficient learning from demonstrations in robotics, particularly for soft object manipulation. Before starting my PhD, I received both my Bachelor's and Master's degrees in Computational Science and Engineering from ETH Zurich, specializing in robotics. My Master's thesis, conducted at the Computational Robotics Lab under the supervision of Dr. Bernhard Thomaszewski and Dr. Roi Poranne, with Prof. Stelian Coros as examiner, explored shape simplification as well as geometric and structural optimization. Alongside my academic work, I enjoy outdoor activities and playing the piano, served on the ETHZ Musikplattform board, and founded Discuna, a course forum platform focusing on collaborative reading.
Research Interests
Invited Speaker and Attendee at Symposium for Advanced Int‘l Students
Had a very nice time in Hangzhou where I was invited to the 'Symposium for Advanced Int‘l Students in the Yangtze River Delta Region' and gave some remarks at the closing ceremony. The symposium aims at giving international students a platform to connect and exchange experiences & our latest research.
Paper Accepted at CVPR 2025
Our paper on 'SemAlign3D: Semantic Correspondence between RGB-Images through Aligning 3D Object-Class Representations' has been accepted as poster at CVPR 2025.
Invited Speaker at SJTU Welcome Event for International Students
I had the opportunity to give a speech at the SJTU Welcome Event for international students.
SemAlign3D: Semantic Correspondence between RGB-Images through Aligning 3D Object-Class Representations
We propose a novel approach for establishing semantic correspondence between RGB images by aligning learnt 3D object-class priors with object instances at inference, reaching SOTA on the challenging SPair-71k dataset.
Towards Simultaneous Topology and Form Finding
This work explored simultaneous planar shape approximation and structural optimization for architectural design. In collaboration with Gramazio Kohler Research, our method was used to design Semiramis, which is an architectural installation at Tech Cluster Zug. A follow-up work, which extended the same underlying methodology to spherical shape approximation, was published at SIGGRAPH 2022.
Advanced Finite Element Simulation for Viscoelastic Foam
Improved soft foam simulation by implementing a higher-order time integration scheme (BDF2) and employing the Rayleigh damping model. This enabled accurate identification of the foam material's parameters by comparing the oscillation behavior of the simulated foam with real-world measurements obtained from a motion capture system.
Discuna
Active DevelopmentCourse Forum focusing on Collaborative Reading.
Discuna uniquely merges the ideas of course forums and collaborative reading to improve student's comprehension of course materials. By combining these two concepts, Discuna helps to declutter notifications and enables an overall more efficient workflow.
“I use Discuna as a communication and feedback tool for three courses for students of Computational Science and Engineering at ETH Zurich. Through Discuna chat channels, I collect questions and suggestions for discussion in the classroom, and through Discuna's PDF annotations feature, the students report errors and issues in the course material. The students very much appreciate the ease with which text and graphics can be blended in Discuna, and I value the clear presentation of communication threads and efficient navigation in Discuna.”
Product Screenshots
Course Forum via Text Channels
You can create text channels for various topics such as feature requests, error reporting, or general Q&A.
Collaborative Reading via PDF Channels
PDF channels allow students to share comments directly on the course materials.
Properly Organized Notifications
Circular badges show the number of unread messages for each channel or folder, and optionally, rectangular badges can display the number of un-/solved issues.
Some of the happiest moments from my journey in research and life that I am incredibly thankful for and will cherish forever.
Performed with the awesome CVPR band at CVPR 2025! 🎸🎤🎹
Reached Mount Fuji at sunrise! 🗻🌄
Trained for several weeks for the Dragon Boat Competition in Shanghai - an unforgettable experience and a great team, many memories and very good friends made! 🐉
Had an amazing evening performing Rachmaninov Prelude Op. 32 No. 2 at the SJTU piano concert! 🎹