With the acceleration of agricultural modernization, agricultural robots are playing an increasingly important role in improving agricultural production efficiency and alleviating labor shortages. However, the complex and changeable agricultural operating environment, such as the different growth forms of crops and changes in the light of the operating environment, greatly limits the autonomy and adaptability of agricultural robots. This paper focuses on this issue by introducing advanced environmental perception algorithms, adapting the robot hardware, building a multimodal data fusion perception system, and optimizing the decision-making control strategy. Experiments show that compared with the traditional A algorithm, the improved adaptive path planning algorithm has a slightly longer path, but it reduces energy consumption, improves the obstacle avoidance success rate, can extend the operation time of agricultural machinery, and reduce the wear rate of mechanical parts. In addition, the algorithm proposed in this paper significantly improves the autonomous operation ability of agricultural robots in complex agricultural scenes, effectively reduces the operation error rate, reduces energy consumption by 14.3%, and increases the obstacle avoidance success rate to 100%, verifying the effectiveness and practicality of the algorithm.