Aiming at the problem of luggage packing in airports where luggage size information cannot be known in advance due to the disordered arrival of luggage with passengers, an online luggage packing method based on reinforcement learning is proposed with the utilization of the packing space of the luggage trolley as the optimization goal. First, a mathematical model of luggage packing was established according to the actual situation of luggage packing at the airport. Then, aiming at the problem of finding a suitable packing position and posture for luggage in the luggage cart, a method for selecting luggage packing position and evaluating method for packing posture was designed. Finally, with the help of the "trial and error" learning mode of reinforcement learning, the online luggage packing strategy was obtained by training the luggage packing model. In simulation experiments, the utility rate of space in baggage cart can reach 82.9%, and the calculation time was 0.39s. Both were better than machine learning algorithms. It has good practicability in solving the airport luggage online packing problem.