Existing self-driving itinerary planning systems rely on single-objective optimization and static cost estimation, limiting their ability to accommodate diverse user preferences and dynamic price fluctuations. Methods: This study presents an interactive itinerary planning system formulated as a Tourist Trip Design Problem (TTDP) with subset selection, integrating an improved NSGA-II algorithm for simultaneous optimization of travel time, cost, and experiential quality, a WPGA-based combined prediction model for dynamic cost forecasting, and a human-in-the-loop interface for real-time preference adjustment. Results: Evaluated on 30 Hainan Island attractions under a 7-day/6-night scenario with 30 independent runs, the improved NSGA-II selects 8.9 ± 1.0 stops with total time 26.2 ± 4.7 h, cost CNY 1526 ± 356, and experience index 36.69 ± 5.60, with backend optimization latency 0.5 ± 0.1 s (end-to-end user-perceived latency including network and frontend rendering: 1.8 s). Compared with standard NSGA-II, the improved version achieves 114.0% higher aggregate experience index while selecting more stops (8.9 vs. 4.2); this improvement is primarily driven by the larger number of selected stops, and normalized indicators (experience per stop: 4.12 vs. 4.08) provide a more balanced interpretation of the trade-offs. The cost model covers three categories (fuel, tickets, and hotel), with the optimization objective f2 capturing route-dependent costs (fuel + tickets) and the hotel treated as a fixed baseline. However, empirical validation is restricted to two geographic corridors.