Mustafa Khan
University of Toronto. Engineering Science.
I’m a 4th-year Engineering Science student at the University of Toronto, majoring in Robotics with a minor in Machine Intelligence.
I invented AutoSplat during a 16-month research internship at Huawei, Noah’s Ark Lab, creating digital twins of autonomous driving scenes using 3D Gaussian Splatting (3DGS). The framework outperformed state-of-the art methods such as EmerNeRF from NVIDIA, leading to a patent and first-author paper submitted to ICRA 2025.
Currently, I’m working on generalizable 3D vision for autonomous driving as part of my bachelor’s thesis with Professor Steven Waslander at the Toronto Robotics and AI Lab (TRAIL).
At the same time, I’m researching the mechanistic interpretability of foundation models (e.g. DINOv2, LLaMA) with Professor Vardan Papyan at the Vector Institute.
I also serve as the Technical Director of Perception at aUToronto, UofT’s self-driving car team, where I lead teams in 2D Vision, 3D Object Detection, and Tracking. Our efforts have propelled the University of Toronto to multiple 1st-place finishes in the SAE Autodrive Challenge against 9 universities.
Research Interests
- Representation learning and interpretability of foundation models
- Self-supervised learning (SSL) and emergence of robust visual features
- Deep learning theory (neural collapse, phase transitions, information bottleneck)
- Generalizable understanding of the world (3D representation learning, vision for autonomous systems)
This site is a work in progress, please excuse the dust.