The realization techniques and current development of vision sensing system of agricultural robot are discussed, which include techniques of distance inspection, characteristic recognition of agricultural production and movement navigation of agricultural robot based on vision sensing. The distance inspection method based on binocular telemeter was used for picking grape or tomato by robots. The distance inspection systems with laser or infrared beam lights were proved which can help to counteract disturbance of sunlight and obtain accurate results. The geometrical shape, gray level, color, and reflected spectrum especially of farm production, can be used for agricultural robots to recognize the operated objects. The navigation method based on artificial guide line is a simple one for agricultural robot navigation, and the array characteristics of plant are also used for agricultural robot navigation. The future development trend of vision sensing system of agricultural robot is also discussed.