This work seeks virtual constraints, or outputs, that are intrinsic to human walking and utilises these outputs to construct controllers which produce human-like bipedal robotic walking. Beginning with experimental human walking data, human outputs are sought, i.e., functions of the kinematics of the human over time, which provides a low-dimensional representation of human walking. It will be shown that, for these outputs, humans act like linear mass-spring-dampers; this yields a time representation of the human outputs through canonical walking functions. Combining these formulations leads to human-inspired controllers that, when utilised in an optimisation problem, provably result in robotic walking that is as ‘human-like’ as possible. This human-inspired approach is applied to multiple human output combinations, from which it is determined which output combination results in the most human-like walking for a robotic model with mean human parameters.