Aiming at the problems of difficult cooperative control and low working efficiency of agricultural Agent groups, the task assignment of agricultural heterogeneous Agent groups was researched based on improved stimulus response model. A layered hybrid multi-Agent architecture based on acquaintance net and the cloud platform-edge server collaborative computing system was established. The stimulus response model of ant colony algorithm was applied to the traditional contract network algorithm, and the adaptive bidding strategy was established to limit the number of bidding Agents and reduce the communication burden of the system. Based on the heterogeneity of agricultural Agents, the efficiency model of task assignment was established, by constructing time-varying coefficient and time matrix, the dynamic trust function and response threshold design method based on direct trust and recommendation-based trust were established to optimize the overall efficiency of agricultural Agent groups. Through increment PID algorithm and integral separated threshold, the adaptive stimulus update function was established to reduce the number of iterations, which reduced the workload of the Agent team overshoot, traffic and the number of iterations when the deviance was converged. The simulation results showed that when the Agent team size was 40 and 100 respectively, the overall efficiency of the improved contract network algorithm was 41.1% and 83.1% higher than that of the traditional contract network algorithm. When the Agent team size was 40, three sets of stimulus update functions were set in addition. The workload overshoot of the stimulus update function based on PID algorithm was reduced by 24.5% and 9.5% respectively compared with the second group and the third group. In terms of iteration times, it was reduced by 84.2% and 84.8% compared with the first group and the third group. When the Agent team size was 20, 40 and 100 respectively, the traffic of the improved contract network algorithm was reduced by 49.1%, 63.7% and 72.4% compared with the traditional contract network algorithm. Experimental verification showed that the traffic and workload overshoot of task allocation by the improved contract net algorithm was reduced by 70.0% and 20.2% compared with the traditional contract net algorithm, the overall efficiency was increased by 14.1% compared with the traditional contract net algorithm, and improved task allocation algorithm could guarantee that the Agent groups at work could achieve full coverage of the work area within the prescribed time limits.