Tag
This paper formalizes the sim-to-real gap for foundation model agents as a Markov Decision Process problem, proposing a unified research agenda to adapt classical solutions like domain randomization for improving agent robustness and reliability in real-world deployment.