We consider a Minimum-Latency Aggregation Scheduling problem in wireless sensor networks when aggregated data from all sensors are required to be transferred to the sink. During one time slot (time is discrete) each sensor can either send or receive one message or be idle. Moreover, only one message should be sent by each sensor during the aggregation session, and the conflicts caused by interference of radio waves must be excluded. It is required to find a min-length conflict-free schedule for transmitting messages along the arcs of the desired spanning aggregation tree (AT) with the root in the sink. This problem is NP-hard in a general case, and also remains NP-hard in a case when AT is given. In this paper, we present a new heuristic algorithm that uses a genetic algorithm and contains the local search procedures and the randomized mutation procedure. The extensive simulation demonstrates a superiority of our algorithm over the best of the previous approaches.