Jan Blumenkamp
Papers
12
Total Citations
86
H-Index
4
About
Jan Blumenkamp is a robotics and artificial intelligence researcher whose work sits at the intersection of multi-robot systems, multi-agent reinforcement learning, and decentralized intelligence. His research addresses one of the field's most persistent challenges: enabling teams of robots to coordinate effectively, scalably, and in real-world conditions without relying on centralized control. Blumenkamp's most influential contribution — a framework for deploying decentralized Graph Neural Network-based policies on real robot swarms (44 citations) — demonstrated that sophisticated learned behaviors could bridge the gap between simulation and physical deployment. Complementing this, he co-developed VMAS, a vectorized multi-agent simulator designed to accelerate collective robot learning research, and explored how offline-optimal expert knowledge can be distilled into online-scalable multi-robot planning policies. More recently, he has pushed the frontier further by integrating Large Language Models into multi-robot navigation through offline reinforcement learning, allowing robot teams to interpret and act on natural language instructions with minimal training data. Across his portfolio, Blumenkamp consistently tackles real-world deployment challenges — from sim-to-real transfer and cooperative perception to decentralized visual SLAM — making him a distinctive voice in building multi-robot systems that genuinely work beyond the laboratory.
Research Focus
Key Achievements
Top Papers
- 1
- 2VMAS: A Vectorized Multi-agent Simulator for Collective Robot Learning12 citations · 2024
- 3
- 4
- 5Language-Conditioned Offline RL for Multi-Robot Navigation3 citations · 2025
- 6
- 7VMAS: A Vectorized Multi-Agent Simulator for Collective Robot Learning3 citations · 2022
- 8
- 9
- 10Closing the Reality Gap with Unsupervised Sim-to-Real Image Translation2 citations · 2022