“…The pick-and-place task showed the capacity of the proposed learning and optimization framework, which can be directly applied to other loco-manipulation scenarios, for example, the door-opening task in [13]. Although we evaluated the demonstrated pick-and-place skills on MOCA, unlike the IRM-based methods for a specific mobile manipulator, it is possible to apply the learned skills to other MMs, such as TIAGo++ in [6], PR2 in [13], NAO humanoid in [9], and even legged MMs [7], as long as the learned EE trajectory is achievable for the MM. Furthermore, owing to the hierarchical design in the proposed HQP, the learned EE trajectory was tracked accurately, and the base pose was followed as a secondary priority, which allowed the transfer of the human whole-body mobile manipulation skills to a robot with different geometry.…”