Sander Tonkens

I am a Ph.D. candidate at UC San Diego advised by Prof. Sylvia Herbert, where I am part of the Contextual Robotics Institute.

During my Ph.D., I have interned at NVIDIA in the Autonomous Vehicle Research Group, working on endowing learned prediction and planning autonomy stacks with interpretable latent representations for explainability and LLM-based steering.

Before moving to San Diego, I did my M.S. at Stanford University, where I was advised by Prof. Marco Pavone. Before that, I was an undergrad at EPFL in Lausanne, Switzerland. I'm currently an Office of Naval Research Naval Innovation, Science, and Engineering Center (ONR NISEC) Fellow. I've previously received the Netherlands America Foundation Graduate Fellowship and am a Fulbright Graduate Scholar.

Email  /  Google Scholar  /  Github

I am currently doing a research stay at Stanford University, where I am working on automated interpretability of failures in robot foundation models for red-teaming.

My research interests lie at the intersection of robotics, control theory, and machine learning. Specifically, I focus on (i) providing probabilistic safety assurances for autonomous systems in unknown, dynamic environments via uncertainty quantification, (ii) infusing interpretability into learned policies for transparent decision-making, and (iii) developing safety filters for OOD runtime monitoring and failure prevention.

[arXiv] [project page]

[arXiv]

[Paper] [video]

[project page] [arXiv] [1 min video] [poster] [RSS'25 OOD workshop paper]

[arXiv] [video]

[project page] [arXiv]

[arXiv] [code] [slides] [poster]

[project page] [arXiv] [code]

[arXiv] [code] [slides] [1 min video]

[arXiv] [code]

[arXiv] [code]