<p>With the rapid advancement of advanced manufacturing and smart logistics, the collaborative optimization and integrated scheduling of production, transportation, and assembly have emerged as a core competency for modern manufacturing enterprises. This article studies a significant transportation-assembly collaborative optimization problem (TACOP), which aims primarily to minimize cycle time, with secondary objectives of minimizing both transportation cost and inventory levels. Since TACOP comprises two coupled sub-problems, namely, the vehicle routing problem (VRP) and the simple assembly line balancing problem (SALBP), a decomposition-based <i>Q</i>-learning evolutionary algorithm (DQEA) is proposed to deal with the TACOP. First, the assembly agent solves the SALBP to acquire an assembly scheme with a minimized cycle time. Subsequently, this assembly scheme is extended into a diverse set of assembly solutions that share the same cycle time but differ in task arrangements. Next, these assembly solutions are decomposed by a slicing method of load balancing to generate specific subproblems, which also facilitates the reduction of inventory levels. Furthermore, given the independence of these VRPs, the transportation agent can solve them in parallel, enabling the efficient derivation of corresponding transportation solutions. To tackle the SALBP and VRPs, the agent utilizes an off-policy approach to adaptively perform global and local search, thereby extracting valuable insights from high-quality actions and rapidly guiding the search direction toward promising regions. Extensive experiments and computational comparisons are carried out to demonstrate the effectiveness and efficiency of DQEA in solving TACOP.</p>