Robert Müller

About

Robert is a reinforcement learning researcher and founder working at the intersection of RL, agentic AI, and self improving systems. His work focuses on AI systems that do not remain static after deployment, but improve through interaction, feedback, traces, and real world outcomes. Before joining the Agentic Systems Lab, Robert worked on deep reinforcement learning for high speed robotic control at Sony AI and AI driven control for electron microscopy at appliedAI. As founding researcher he build the research team at Convergence, the AI agent startup later acquired by Salesforce. His research spans policy gradient methods, meta learning across task distributions, multi agent environments, and strategic interaction between language model agents, including recent work on agent benchmarks for negotiation and imperfect information settings. Robert is the founder of Aganthos, where he works on learning from experience as a foundation for self improving AI systems. The company explores how single agents and multi agent workflows can learn from accumulated experience, adapt their behavior over time, and become more capable through use. At ASL, Robert focuses on reinforcement learning for self improving agents: systems that can evaluate their own behavior, extract useful learning signals from logs and feedback, and improve prompts, tools, routing decisions, and model weights over time. His broader ambition is to connect the empirical power of foundation models with the adaptive machinery of reinforcement learning.