In big data-driven new retail, the reconstruction of “people, goods, marketplaces” has fostered online-offline integration and intelligent services, spurring a surge in consumers’ demand for “instant delivery”. However, traditional logistics faces three core bottlenecks: static capacity scheduling fails to adapt to urban traffic’s spatiotemporal dynamics, such as peak-hour UAV takeoff/landing congestion and redundant truck stays; short-distance delivery relies on the “UAV-truck-UAV” transfer link, causing poor timeliness; and no unified framework quantifies transportation, time, and carbon emission costs, hindering multi-objective optimization balance. To address these, this paper proposes a truck-UAV joint distribution route optimization model with dynamic recycling. Methodologically, it builds a “time-phased & zone-divided” dynamic recycling framework where UAVs operate at dedicated points during peak hours and return to original trucks during off-peak periods; develops a UAV short-distance direct delivery mode that eliminates truck transfers for merchant-customer short-distance orders; introduces dynamic battery consumption functions and time-sensitive coefficients to quantify the three costs; and designs a hybrid heuristic algorithm integrated with machine learning for screening high-quality solutions and deep learning for optimizing initial populations to solve the “task-route” combinatorial explosion in dynamic scenarios. The main contributions include overcoming traditional static recycling limits to adapt to new retail’s temporal variability; simplifying short-distance links to meet instantaneity demands; establishing a unified multi-cost quantification model that transforms multi-objective optimization into weighted total costs minimization.