QUAD-AV - Problem
The US National Academy of Engineering recognized agriculturalmechanization as the 7th greatest engineering achievement of the 20th century, of which other top ten achievements included electrification, automobile, and airplane (Constable and Somerville, 2003). This is a clear recognition of the significance of farm machinery that transformed traditional agriculture to an industry and greatly increased its efficiency and productivity.
Looking ahead, agricultural automation could be one of the greatest engineering achievements of the 21st century. One specific field of automation is the autonomous navigation of tractors and implements on the field, which can lead to several advantages. One advantage is the saving of human operator time (see e.g. Zhang et al., 2010; Happich, Lang & Harms, 2010), another is facilitation of more efficient farming methods, for instance due to regular monitoring of plant growth and precision type of plant nursing (see e.g. Klose et al., 2010).
Autonomous navigation requires path-planning, which again is dependent on self-localization. For self-localization, the most common approach is the use of a Global Navigation Satellite System (GNSS) like the well-known GPS or the future European Galileo. The quality of such a system can be enhanced by a network of RTK stations, which provides high-accuracy and high-reliability coordinate corrections via various communication channels and protocols (Keenan, 2010). Autoguided agricultural machines based on this approach are already in practical use for some years.
An example of a GPS-based auto-guidance system is John Deere's AutoTrac system (Kise et al., 2010). Such auto-steering systems can reduce the overlap of adjacent paths and reduce operators' fatigue, further improvement on the productivity and efficiency can be achieved.
Despite all success with GPS and RTK based systems, which allow for precise navigation of agricultural vehicles, these sensors do not however provide any information on the dynamics of the environment. For instance, in agricultural applications one can often assume that there already exists a coarse map of the terrain where the autonomous vehicle is operating. However, we cannot blindly trust this map, because it might contain errors due to recent changes in the field caused by men or by nature. Beyond that, the map does not contain any information about moving objects (human beings, animals and vehicles) that actually might reside in the terrain.
The later fact poses a major safety issue related to any type of autonomous navigation and operation. The challenge is now to develop smarter, fully autonomous vehicles able to operate safely in semi-structured or unstructured dynamic environments, in which also humans and animals may be present. In this respect, the safety aspect of autonomy is one the most critical issues.
Safety should be addressed not only to prevent the vehicle from getting damaged, but also to guarantee the security of people, animals, and goods. To overcome this important challenge, researchers need to provide the vehicle with a clear understanding of its surroundings; the cognitive ability to perceive the environment is in many cases a matter of saving thousands of € worth of machine or yield damages.