Mitarbeitende

Weed control is crucial to attain high productivity and quality in field crops. The application of chemical herbicides has been the major strategy to control weeds during the last decades in most European countries. However, herbicides can have negative impacts on the environment and traces of herbicides can be found in food and in surface and ground water. Public concerns due to these negative impacts have led to an increasingly political pressure to reduce herbicide use. In 2016, Agroscope, the Swiss federal center for agricultural research, started a project to support the development of an autonomous weeding robot for the selective and precise application of herbicides. The first activities consisted in training an algorithm to recognize weeds and in 2017 the first field trials were conducted. The objective of this contribution is to share first experiences based on preliminary results obtained from experiments with sugar beet and oilseed rape. The efficacy of the robot to control weeds was monitored in areas of 2 m2 in a field cultivated with sugar beet. Additional, experiments with spring and winter oilseed rape were conducted in plots of 13.5 m2 and weeds and volunteers were monitored in subplots of 0.25 m2. Oilseed rape and quinoa (Chenopodium quinoa) were planted perpendicularly to the crop rows to simulate volunteers and Chenopodium sp. weeds, respectively. In the experiments with Oilseed rape the efficacy of the robot was assessed for three different strategies to control weeds: i) application of a selective herbicide after crop emergence, ii) application of a pre-emergence herbicide and a broad spectrum post-emergence herbicide, and iii) post-emergence application of a broad spectrum herbicide. In addition, the experiment included treatments with no-weed control and three additional weed control treatments: i) conventional post-emergence control, ii) mechanical control and iii) biological control using a non-winter hardy cover crop. The rate of recognition of weeds and volunteers by the robot was above 60% for specific treatments which suggests a promising potential for autonomous weeding robots to become useful tools for integrated weed management systems. The fact that recognition efficacy depended on weed or volunteer density suggests that different robot settings and utilization strategies need to be considered according to densities of weeds and volunteers. Agronomic research is needed to optimize the use of weeding robots and to properly integrate them into cropping systems.