The integration of autonomous robots in agriculture presents significant potential for optimizing post-harvest crop collection, addressing challenges such as labor shortages, operational inefficiencies, and scalability in large-scale farming. This paper introduces a novel optimization model for the Crop Collection Problem, formulated as a Flexible Multi-Depot Capacitated Vehicle Routing Pickup Problem (FMDCVRP-P). The proposed model coordinates a fleet of autonomous robots to collect harvested crops from multiple field locations and deliver them to various depots. By allowing flexible depot assignments, the model minimizes unnecessary travel, enhancing operational efficiency compared to traditional fixed-depot approaches. The FMDCVRP-P is formulated as a mixed-integer linear programming (MILP) problem, aiming to minimize the total makespan. Numerical experiments demonstrate that flexible depot assignments significantly improve crop collection efficiency. Additionally, a real-world robotic demonstration validates the model, showcasing its feasibility and effectiveness in practical scenarios.