Humanoid-robots are increasingly relevant to industrial production, yet deployment is limited by the sim-to-real gap - the difficulty of transferring control and perception policies trained in simulation to real factories. We propose a Formal Conceptual Framework (FCF) to reduce this gap and accelerate technology transfer to aerospace manufacturing. The FCF integrates three pillars: domain randomization (DR) to build robustness, whole-body control (WBC) to ensure dynamic feasibility and balance, and hybrid reinforcement/imitation learning (RL/IL) to leverage demonstrations while optimizing performance. As a case study, we consider assembly and inspection of aircraft fuselage components, a task demanding precision, stability, and safe human robot collaboration. Our contribution is a unified sim-to-real validation loop that couples DR-trained policies with optimization-based WBC under privileged learning, together with a reproducible, laboratory-backed workflow for iterative data collection, system identification, and simulator updating. The AI and Humanoid-Robots Laboratory at Rzeszow University of Technology provides the testbed for this loop, enabling quantitative evaluation of robustness, success rate, and safety across repeated sim ↔ real cycles.
Goral et al. (Wed,) studied this question.