The Atlas motion platform was conceptually introduced in 2005 as a 2.90 m diameter thin-walled composite sphere housing a cockpit. Three active mecanum wheels provide three linearly independent torque inputs enabling the sphere to enjoy a 100% dexterous reachable workspace with unbounded rotations about any axis. Three linearly independent translations of the sphere centre, decoupled from the orientation workspace, are provided by a translational three degree-of-freedom platform. Small-scale and half-scale demonstrators introduced in 2005 and 2009, respectively, gave us the confidence needed to begin the full-scale design. Actuation and control of the Atlas full-scale design is nearing completion; however, resolution of several details have proven extremely elusive. The focus of this paper is on the design path of the 24 passive mecanum wheels. The 12 passive wheels below the equator of the sphere help distribute the static and dynamic loads, while 12 passive wheels above the equator, attached to a pneumatically actuated halo, provide sufficient downward force so that the normal force between the three active wheel contact patches and sphere surface enable effective torque transfer. This paper details the issues associated with the original twin-hub passive wheels and the resolution of those issues with the current split-axle design. Results of static and dynamic load tests are discussed.