Container transport requires real-time control and a high degree of cooperation to alleviate disturbances and perform smoothly without unnecessary environmental impact and monetary losses. The involved operators are, however, often reluctant to cooperate as they fear loosing valuable information and autonomy, eventually leading their clients to choose another (cooperating) operator. In this paper we propose a real-time co-planning method, called Secure Departure Learning that in real-time lets several truck operators indicate to a barge operator what departure schedules they prefer without revealing any sensitive information. The method uses Paillier encryption and a learning method inspired by Bayesian Optimization in a model predictive control framework. At frequent time intervals a number of potential barge schedules are communicated to the co-planning truck operators. They evaluate their operation cost for each schedule and communicate it to the barge operator encrypted using several public keys. The barge operator computes the encrypted total cost for each schedules, which hereafter is decrypted by several truck operators. The first action of the schedule that results in the lowest total cost is implemented in a model predictive control fashion. Simulated experiments on a realistic, Dutch transport network illustrate that Secure Departure Learning is a good alternative for replacing the current method in practice, where barge departures are scheduled ahead of time and only mode-decisions can be updated in real time. Secure Departure Learning offers a new perspective on cooperation at the operational level in freight transport where co-planning and information protection can go hand in hand.