Improving robustness of walking robots has always been problematic. Their complex kinematics and locomotion has always been prone to damage: a broken cable, an unstable foothold or a wrong set of parameters has been an everlasting source of frustration. Nature developed an extraordinary robustness through redundancy and fast adaptation. Theories about decentralized nervous systems has inspired this paper with a novel approach. The presented solution aims at relocating low-level walking behaviours to a network of computers and, more exactly, into the robots individual legs. This paper will not cover the full scope of the software implementation (this is a eld found especially in modular robotics), but presents how such an encapsulated leg with all necessary hardware is built and focuses on the mechanical and kinematic aspect of such legs. It highlights how a robotic leg needs to be designed to tackle structured environments serves as explanatory guide through the design process of legs with integrated PCU and sensors.