Research

Research Overview

My research lies at the intersection of reinforcement learning, control theory, and AI safety. I focus on developing theoretically grounded algorithms for sequential decision-making problems with complex objectives, particularly in safety-critical applications.

🎯 Submodular Reinforcement Learning

Bridging submodular optimization with RL to handle non-additive reward structures that arise in real-world applications like experiment design and informative path planning.

Submodularity Non-Markovian Rewards Approximation Theory

🛡️ Safe Multi-Agent Learning

Developing provably safe algorithms for multi-agent systems with applications in coverage control, autonomous racing, and distributed robotics.

Safety Guarantees Multi-Agent RL Coverage Control

🏁 Competitive Policy Optimization

Novel approaches to multi-agent reinforcement learning through competitive optimization, with applications in autonomous racing and game-theoretic scenarios.

Game Theory Policy Optimization Autonomous Racing

🤖 Model Predictive Control

Integrating learning with control theory for safe autonomous systems, focusing on Gaussian processes and Bayesian optimization for uncertainty quantification.

MPC Gaussian Processes Bayesian Optimization

Current Projects

🔬 Submodular Reinforcement Learning

Collaborators: Mojmir Mutny, Melanie Zeilinger, Andreas Krause

We investigate reinforcement learning problems where the reward function exhibits submodular structure. While these problems are NP-hard to approximate, we develop practical algorithms with conditional approximation guarantees. Applications include:

Experiment design for scientific discovery
Informative path planning for environmental monitoring
Molecular design with diversity constraints

Status: Under review at top-tier venue

🚗 Safe Coverage Control for Multi-Agent Systems

Collaborators: Matteo Turchetta, Melanie Zeilinger, Andreas Krause

This work addresses the challenge of coordinating multiple agents to cover a spatial domain while maintaining safety guarantees. We provide near-optimal algorithms with theoretical guarantees.

Published: NeurIPS 2022

🏎️ Autonomous Racing Systems

Collaborators: AMZ Racing Team, ETH Zurich

Development of perception, planning, and control algorithms for autonomous race cars. Our system won the Formula Student Germany and Italy championships in 2018.

Published: Journal of Field Robotics 2020

Research Philosophy

🎯 Theory-Practice Bridge

I believe in developing algorithms that are both theoretically principled and practically implementable. Every theoretical contribution should have clear practical implications.

🛡️ Safety First

Safety is not an afterthought but a fundamental design principle. I focus on provable safety guarantees for autonomous systems operating in uncertain environments.

🌐 Interdisciplinary Approach

Complex real-world problems require interdisciplinary solutions. I draw from control theory, optimization, machine learning, and robotics to tackle challenging problems.

Collaborations

I’m fortunate to collaborate with leading researchers across multiple institutions:

ETH Zurich: Andreas Krause, Melanie Zeilinger, Matteo Turchetta
Caltech: Yisong Yue, Anima Anandkumar, Kamyar Azizzadenesheli
Industry: Fixposition AG, AMZ Racing

Future Directions

Looking ahead, I’m excited about several emerging research directions:

Foundation Models for Control: Integrating large language models with control systems
Quantum-Enhanced RL: Exploring quantum computing applications in reinforcement learning
Sustainable AI: Developing energy-efficient algorithms for autonomous systems
Human-AI Collaboration: Safe interaction between humans and autonomous agents

Interested in collaboration? Feel free to reach out!