Collaborative project: Electronic towing bar for agricultural machines with surroundings sensor system and adoption of geo-referenced information – Subproject 2 | FISA - Forschungsinformationssystem Agrar und Ernährung

Collaborative project: Electronic towing bar for agricultural machines with surroundings sensor system and adoption of geo-referenced information – Subproject 2

Project

Project code: 2815306707
Contract period: 01.03.2011 - 31.07.2014
Budget: 201,969 Euro
Purpose of research: Experimental development

Increasing cost pressure on the global agricultural market forces the farming industry to continuously improve productivity and cost efficiency. The consolidation of small acres and the use of more powerful machines provide better economies of scale. Electronic assistance systems for process optimization and automation have gained importance. The electronic tow bar concept enables the simultaneous operation of two tractors on site with only one driver. Both vehicles are equipped with high precision GNSS receivers and are connected via a RF data link exchanging process, system and position data. The unmanned slave tractor follows the manned master tractor maintaining a preset lateral and longitudinal offset while copying the process parameters from the master. The functional concept of the electronic tow bar system has been developed and tested in a preceding research project (EDA). Based on these results, the EDAUG project has focused and advanced the aspects safety and usability by integrating of Environment perception sensors and geo-information into the tow bar concept. To avoid collisions, environment perception sensors gather data about the close vehicle environment of the slave tractor and indicate the intrusion of obstacles on the planned path into a safety relevant area. A positive event triggers a stopping maneuver of the vehicle before the collision event (Chair for Mobile Machines, KIT). Geo-Information concerning the actual field are downloaded on site via a mobile internet connection. The data package contains a-priori information about stationary, previously known obstacles. This knowledge enables the proactive generation of a collision evasion path without interruption of the process (geo-konzept GmbH). A sophisticated HMI concept and a system level analysis of functional safety aspects have been contributed by the AGCO GmbH. The environment perception sensor concept for the slave tractor consists of two 2D-LIDAR sensors and four 3D-ToF cameras to monitor the relevant hazard area. 3D-ToF cameras deliver a 3D-point cloud up to a distance of 5m. The LIDAR sensors are heading forward seamlessly connecting to the 3D-camera FOV and cover a range up to 80m. To compensate an uneven ground topology, the scanning plane is continuously leveled in two axes to guarantee the detection of a minimum obstacle height at a critical distance ahead. The elevation (pitch) is controlled by a speed dependent non-linear control algorithm, while the lateral (roll) angle of the scanning plane is controlled to maintain a rectangular scanning plane/ground intersection line referred to the tractor heading. At zero speed, a robust differential image algorithm produces the 3D point cloud segmentation. When moving forward, a grid based iterative total least square fitting delivers a ground plane estimation and the obstacle segmentation. The subdivision of the environment into a grid enables the smooth fitting of the ground hypotheses to uneven ground topology and hence reduces segmentation errors due to bumps and dips. The 2D-LIDAR data is processed into a histogram of ground plane parameters. The best ground plane estimation is used to segment the data into ground and obstacle points. The suitability of the developed algorithms could be validated in simulations and with real test data. However, the chosen 3D-cameras have been too sensitive to outdoor backlight conditions and ego-motion of the vehicle due to motion artefacts. To allocate geo-information data sets for use on the tow-bar platoons, a server has been implemented that accesses and bundles geo-information from open source and commercial web services. The data is downloaded to the slave tractor on site via a mobile internet connection to generate an obstacle list. The contour and classification of the obstacles and the actual field boundary is transmitted to the master to be visualized on the HMI. To evade a-priori known obstacles, a trajectory based approach has been implemented and successfully tested in a final field test. The overall system behavior is controlled via a central state machine, which runs on master and slave tractor synchronously. The state machine features the three-level safety concept consisting of the safety states Operational, Safety-Stop and Emergency-Stop. Concerning the hardware architecture, the state machine is a central node of the data communication. Hence, all relevant parameters, error messages, and communication paths can be monitored and evaluated for the decision making of the overall system state. If on one vehicle a transition condition to a save state becomes true, the prompt transition execution follows. The new vehicle state is then communicated to the other vehicle and copied immediately. In consequence of the central position, the state machine ECU functions as well as network gateway between the RF-connection and the local CAN bus of the tow-bar system. A remaining bus simulation has been used, to validate the software functionality and data integrity previously to the field tests. To handle high bandwidth of environment modelling data, Ethernet has been implemented between the concerned ECUs additionally to the CAN network. The functionality, operability and safety of the system concept have been validated in field tests with a prototype platoon. The environment perception concept has been validated separately. The HMI concept integrates all process, machine and environment information presenting a reduced to the most urgent set of data to the driver. Intuitive navigation, efficient use and the tailored information allocation preventing information overflow while ensuring good system behavior traceability have been the main drivers for the HMI development. The environment model containing obstacles, field boundaries and path planning is visualized in the map of the standard Vario terminal of the tractors. Each tracked obstacle shown in the map can be selected to determine the further treatment. The existence of obstacles can be confirmed or disproved and an evasion path or stopping maneuver can be allowed or ignored. If no driver input has been traced, the default maneuver is to approach an obstacle as close as safely possible and then trigger a Safety-Stop. The EDAUG project has produced a solution to operate a semi-autonomous tractor platoon on the field that features a high integration level of environment data, process automation and HMI menu navigation. The concept has been implemented and validated in functional module tests, simulation and prototype field tests. The results offer valuable experience about the controllability of unmanned vehicles in agricultural processes and delivers a suitable approach how to deal with a high load of information when simultaneously operating multiple machines.

show more show less

Section overview