FroboMind, proposing a conceptual architecture
for agricultural field robot navigation
Kjeld Jensen1, Anders Bøgild2, Søren Hundevadt Nielsen1, Martin Peter Christiansen2, Rasmus Nyholm Jørgensen1
1Institute of Chemical Engineering, Biotechnology and Environmental Technology
2The Maersk Mc-Kinney Moller Institute
University of Southern Denmark, Campusvej 55, 5230 Odense M, Denmark
Corresponding author: kjen@kbm.sdu.dk
Today farmers are experiencing an increasingly competitive market of crop production, and they are constantly trying to improve their competitiveness by increasing the yield while at the same time lowering expenses to labour and machinery. In the later years they are in addition experiencing new requirements from both legislators and consumers towards more sustainable crop production (Charles el al. 2010).
Precision agriculture utilizing autonomous field robotic vehicles to perform plant nursing tasks efficient and reliable without human intervention has the potential to improve competitiveness and increase sustainability (Chaoui & Sørensen, 2008), (Blackmore et al. 2005). This has pushed the development of autonomous tractor guidance systems and field robotics tool carriers for many years now.
A number of field robotic prototypes have been developed, but state of the art is limited to simple navigation in semi-structured areas like row crops, waypoint navigation or a combination hereof. In addition they do not contain a proper safety system (H.W. Griepentroch et al. 2009). Safety is often simplified to performing an emergency stop when unforeseen events like e.g. a detected moving obstacle occur.
The current development tends to focus on navigational tasks including reliable robot localization with a reduced dependency on RTK-GPS, accurate driving in row crops and orchards, dynamical mapping and route planning, and internal fault diagnosis. The complete autonomous robotic context is described by several other sources of knowledge as well which influences the behavior. Examples are the robot task description, system feedback, implement feedback, static and dynamic obstacle detection, interaction with external collaborating entities etc. (Blackmore et al. 2002).
Moving towards fully autonomous plant nursing robots requires a system architecture that can accommodate these many different components and provide a structured approach to the required exchange of data between them. True autonomous operation additionally requires implementation of a cognitive module within this architecture which processes all relevant knowledge in order to assess possible behaviors and determine the most suitable one for the robot.
The aim of this work is to propose a conceptual system architecture Field Robot Cognitive System Architecture (FroboMind). which can provide the flexibility and extend ability required for further research and development within cognition based navigation of plant nursing robots.
Field Robot Cognitive System Architecture (FroboMind)
FroboMind is based on experience from developing several field robotic platforms at the University of Southern Denmark and Aarhus University, namely Agrobot (Nielsen, 2008), Hortibot (Jørgensen et al. 2007), Casmobot ( and Omnirota (Jørgensen & Maagaard, 2008). The current literature lists various articles proposing different system architectures for field robots. A number of these have been studied, and some of the ideas and concepts have been incorporated into FroboMind.
Looking at the various processes that constitutes the interaction of a field robot with it’s environment, a layered overview like figure 1 may be used.
Figure 1: Generic process overview
Sensing the environment provides data which needs to be processed in order to obtain the information herein. This information is then combined and filtered to extract relevant knowledge, which forms the base for planning the optimal navigation route, implement control etc. The plans are then executed with respect to state and time using commands which are converted to actuation signals through low level controllers. The actuation causes a state change which sources new stimuli and hence closes the loop.
The figure represents a coarse simplification in the sense that in any practical field robot implementation the processes will not be strictly layered. For instance a safety component must able to abort actuation without relying on any process in between, knowledge shared between collaborating entities does not need to be processed and extracted etc.
Figure 2: FroboMind conceptual architecture
Figure 2 shows an expansion of the generic process overview into the FroboMind conceptual architecture. In order not to clutter the overview it is assumed that any component has access to data accessible by it’s predecessor, and therefore multiple connections to successors are shown only when relevant to the understanding of the architecture. Data available for all components like timing and time synchronization, as well as knowledge shared with collaborating entities have not been included in the overview.
Current Project Status
Current status of the project is that a preliminary version of the conceptual architecture has been described and a corresponding software implementation of the architecture is being developed using the Robot Operating System (ROS) (Nielsen et al. 2011). In order to validate the usefulness and flexibility of the conceptual architecture, the software implementation is currently ported to two quite different plant nursing robots:
Figure 3: An early picture of ASuBot fitted with brackets for sensor mounting etc.
ASuBot is joint project between The University of Southern Denmark and Aarhus University. The vehicle is a commercially available garden tractor fitted with a commercial front wheel steering unit. A bracket has been attached to the vehicle for mounting Laser Range Scanner, Stereo camera and GPS-antenna, and position encoders have been mounted on the rear wheels. A detailed description of ASuBot including the software implementation of FroboMind can be found in (Nielsen et al. 2011). Localization in orchards without using GPS is used as a test case for FroboMind component development (Christiansen et al. 2011). This is a small scale test of how the architecture performs when applied as an autonomous guidance system to a standard commercial tractor.
Figure 4: Design drawings of the Armadillo robotic tool carrier.
Armadillo is joint project between The University of Southern Denmark and Aarhus University. It is an electric powered robotic tool carrier to participate in the Field Robot Event 2011 ( in Denmark and to support current research and development projects at the universities. A complete area coverage component for applications like grass mowing and detection of landmine contaminated sites (Jensen et al. 2011) is used as a test case for FroboMind component development (Aslund et al. 2011). This is a test of how the FroboMind architecture performs when implemented as a core operating system for an autonomous robotic tool carrier.
Acknowledgement
This research is linked to and partially funded from the Danish Ministry for Food, Agriculture and Fisheries project: FruitGrowth (Journal No. 3405-10-OP-00146). The authors wish to thank Ole Juul Jørgensen from Aarhus University for the illustrations of Armadillo.
References
Aslund S., Jensen K., Jørgensen R. N. 2011.Completecoveragepathplanningofarandompolygon - AFroboMindcomponent. NJF seminar 441, Automation and System Technology in Plant Production CIGR section V & NJF section VII conference, 30 June - 2 July 2011 Denmark (Publication within this conference).
Blackmore S., Fountas S. and Have H. 2002. A proposed system architecture to enable behavioural control of an autonomous tractor. In: Automation Technology for Off-Road Equipment, Proceedings of the July 26-27, 2002 Conference (Chicago, Illinois, USA). pp. 13-23
Blackmore S., Stout B., Wang M., Runov B. 2005. Robotic Agriculture - Future of the Agricultural Mechanisation? 5th European Conference on Precision Agriculture Uppsala, Sweden 9-12th June 2005.
Christiansen M. P., Jensen K., Jørgensen R. N., Ellekilde L.-P. 2011.LocalizationinorchardsusingExtendedKalmanFilterforsensor-fusion - AFroboMindcomponent. NJF seminar 441, Automation and System Technology in Plant Production CIGR section V & NJF section VII conference, 30 June - 2 July 2011 Denmark (Publication within this conference).
Charles H., Godfray J., Beddington J. R., Crute I. R., Haddad L., Lawrence D., Muir J. F., Pretty J., Robinson S., Thomas S. M., and Toulmin S.. 2010. Food Security: The Challenge of Feeding 9 Billion People. Science 12 February 2010: Vol. 327 no. 5967 pp. 812-818.
Chaoui H. I., Sørensen C. G.. 2008. Review of Technological Advances and Technological Needs in Ecological Agriculture (Organic Farming). ASABE Conference, Providence, Rhode Island, June 29 – July 2, 2008
Griepentrog H.W., Andersen N.A., Andersen J.C., Blanke M., Heinemann O., Madsen T. E., Nielsen J., Pedersen S. M., Ravn O. and Wulfsohn D. 2009. Safe and reliable: further development of a field robot. Precision Agriculture 09.
Jensen K., Jørgensen R. N., Bøgild A., Jørgensen O. J., and Nielsen S. H. 2011, 20 June - 21 June 2011 Belgium, Autonomous mapping of landmine and unexploded ordnance contaminated sites.
Jørgensen, R.N., Sørensen, C.G., Pedersen, J.M., Havn, I., Jensen, K., Søgaard, H.T., Sørensen L.B.. 2007. Hortibot: A system design of a robotic tool carrier for high-tech plant nursing. CIGR Journal of AE Scientific Research and Development, Volume IX, 13 pp
Jørgensen R.N. & Maagaard J. 2008. Omni rotational driving and steering wheel. Patent No. WO2008138346
Nielsen, K. K., Andersen, HJ, Thomsen, A, Jørgensen, JR & Jørgensen, RN 2009, ' Estimation of leaf area index in cereal crops using red–green images ', Biosystems Engineering , vol 104, nr. 3, s. 308-317.
Nielsen S.H., Bøgild A., Jensen K., Jørgensen R.N. 2011.ImplementationsofFroboMindusingtheRobotOperatingSystemframework. NJF seminar 441, Automation and System Technology in Plant Production CIGR section V & NJF section VII conference, 30 June - 2 July 2011 Denmark (Publication within this conference).