Manish Prajapat
Amazon AI Labs | Ph.D. ETH Zurich | Caltech | IIT Madras
I build autonomous AI agents at Amazon AI Labs. Previously, I was a doctoral researcher at the AI Center at ETH Zurich, advised by Andreas Krause and Melanie Zeilinger. My thesis was on reinforcement learning and controls via information theoretic exploration. I focused on sequential decision making under uncertainty arising from world models, constraints or unknown objectives. I developed practical algorithms with theoretical guarantees of safety and optimality.
I received my Master's from ETH Zurich and Bachelor's from IIT Madras. I wrote my Master's thesis at Caltech on multi-agent reinforcement learning with Yisong Yue, Kamyar Azizzadenesheli, and Anima Anandkumar. Before my PhD, I worked as a research engineer at Fixposition AG on sensor fusion for autonomous robots and was part of the AMZ Racing Formula Student Driverless team, which won Formula Student Germany and Italy in 2018.
News
- Jan 2026 Paper on Safe Exploration via Policy Priors accepted at ICLR 2026.
- Dec 2025 Joined Amazon AI Labs working on autonomous AI agents.
- Dec 2025 Successfully defended my Ph.D. at ETH Zurich!
- 2025 Papers accepted at NeurIPS 2025 (SonoGym), TAC 2025, and RA-L 2025.
- 2024 Submodular RL accepted at ICLR 2024 (Spotlight). Best Paper Finalist at CDC 2024.
Selected Publications
Performance-Driven Constrained Optimal Auto-Tuner for MPC
IEEE Robotics and Automation Letters (RA-L) 2025
Safe Guaranteed Exploration for Non-linear Systems
IEEE Transactions on Automatic Control (TAC) 2025
Global Reinforcement Learning: Beyond Linear and Convex Rewards via Submodular Semi-gradient Methods
ICML 2024
Towards Safe and Tractable GP-MPC: Efficient Sampling within a SQP Framework
CDC 2024 (Best Paper Award Finalist)
AMZ Driverless: The Full Autonomous Racing System
Journal of Field Robotics, 2020 (150+ citations)
Blog
Coming Soon
Thoughts on Submodular Reinforcement Learning
An accessible introduction to submodular RL and why non-additive rewards matter for real-world applications.
Read moreComing Soon
Lessons from Autonomous Racing
What building a championship-winning autonomous race car taught me about systems engineering and teamwork.
Read moreComing Soon
Safe Learning in Multi-Agent Systems
The challenges and opportunities in developing provably safe algorithms for multi-agent coordination.
Read moreContact
I'm always happy to discuss research, potential collaborations, or interesting problems in reinforcement learning and AI safety. Feel free to reach out!
manishp@ai.ethz.chSeattle, WA