In autonomous driving, sensing is the most fundamental task providing the necessary information for intelligent vehicles. Compared with single-vehicle sensing, cooperative sensing can greatly reduce the cost, increase the accuracy and overcome the vision range limit. In this paper, a cooperative sensing framework is proposed based upon the occupancy grid map. For map formation, we investigate four occupancy probability distribution algorithms to transform the sensor data into a probability model; for map fusion, we design a probability fusion method to combine multi-vehicle maps. Simulations show that the proposed probability distribution algorithms can capture the environment information with different focuses and the map fusion process can expand the vehicles' sensing range and improve the accuracy.