Jouko Kalmari

ISO 11783 Compliant Forest Crane as a Platform for Automatic Control

Abstract ISO 11783 is a communication standard for agricultural and forest machines. This standard allows an implement to command specific functions of a tractor. Agricultural tractors can be equipped for silvicultural work forming small scale forest machine. It could cost-efficiently compete against common forest machines in some tasks. We have developed an ISO 11783 compliant forest crane connected to an agricultural tractor. The combination is designed to work as a test platform for an autonomous forest machine. The dynamics of the system have been studied using first and second-order models. Based on identification tests with no load on the crane, first-order model is sufficient for describing the motion of most of the cylinders. According to the identification results, small controls do not cause motion on the crane, and a non-linear model is required. Currently used hydraulics of agricultural tractors is not entirely adequate for controlling forest cranes. With more intelligent tractor hydraulics, the crane could be more controllable and energy-efficient.

Sway Estimation Using Inertial Measurement Units for Cranes with a Rotating Tool

Abstract Cranes have often a freely hanging load or tool that starts easily swaying. Anti-sway control requires that the angles and angular velocities of the swinging object are measured. Some cranes can also rotate the tool with a hydraulic motor, and in many cases this rotator angle should also be known. Instrumenting all three axes, two swaying and one rotating axis, with traditional rotary encoders can be challenging. We propose an extended Kalman filter based system using two inertial measurement units. This system can measure the swaying in both directions and estimate the rotator angle. Computer vision system is used as reference. The initial results show that the error is approximately 5 degrees in the rotator angle and 2 degrees in the sway angles. The observer runs at 100 Hz on an embedded microcontroller.

Coordinated motion of a hydraulic forestry crane and a vehicle using nonlinear model predictive control

Coordinated control of a hydraulic forestry crane and a vehicle platform is proposed.The vehicle follows a given path and the crane a time based trajectory.Path and trajectory tracking is based on nonlinear model predictive control.Average tracking error of the vehicle was 6.736.2cm and boom tip 3.713.6cm. Forests are a challenging environment for autonomous operations. The automatic driving and control of hydraulic forestry cranes have been studied earlier as separate problems. However, this study focuses on control where the boom and the forest machine are employed in a coordinated manner. Such control could be beneficial in certain tasks, such as forest cleaning operations where small trees are removed. To accomplish the coordinated actions, nonlinear model predictive control (NMPC) is utilized. NMPC is a control strategy based on numerical optimization which minimizes a given objective function. The coordinated control was tested with real hardware consisting of a tractor and a forestry crane. In the tests, the tractor has a target path and the tip of the boom has a target trajectory. These tests correspond to a real world situation where the forest machine has its own driving lines and the tool is used to accomplish a given task. Two different trajectories for the boom tip were tested with the target velocity of the boom tip being 0.5m/s or 1.0m/s. At these velocities, the average tracking error of the tractor ranged from 6.7cm to 36.2cm while the average error of the boom tip varied between 3.7cm and 13.6cm. In a separate test where only the tractor was controlled, its tracking error was 15.4cm.

Real-time detection of young spruce using color and texture features on an autonomous forest machine

Forest machines are manually operated machines that are efficient when operated by a professional. Point cleaning is a silvicultural task in which weeds are removed around a young spruce tree. To automate point cleaning, machine vision methods are used for identifying spruce trees. A texture analysis method based on the Radon and wavelet transforms is implemented for the task. Real-time GPU implementation of algorithms is programmed using CUDA framework. Compared to a single thread CPU implementation, our GPU implementation is between 18 to 80 times faster depending on the size of image blocks used. Color information is used in addition of texture and a location estimate of the tree is extracted from the detection result. The developed spruce detection system is used as a part of an autonomous point cleaning machine. To control the system, an integrated user interface is presented. It allows the operator to control, monitor and train the system online.

Detection and species classification of young trees using machine perception for a semi-autonomous forest machine

An approach to automatically detect and classify young spruce and birch trees in forest environment is presented. The method could be used in autonomous or semi-autonomous forest machines during tending operations. Detection is done by segmenting laser range images formed by a rotating laser scanner. Classification is done with a two-class Naive Bayes classifier based on image texture features. Multiple combinations of 99 features were tested and the best classifier included eight features from the co-occurrence matrix, local binary patterns, statistical geometrical features and Gabor filter. 79% of spruces and birches in the testing material were detected and 74% of these were correctly classified. Results suggest that the approach is suitable but there are still some challenges in each of the processing steps. Iteration between segmentation and classification is needed to increase reliability.

Measuring Log Length using Computer Stereo Vision in Harvester Head

Abstract Sales of harvested wood is based on harvester head measurements. Therefore, an accurate length measurement is important when a log is being processed on a forest harvester. A stereo camera pair was mounted to the harvester head and the motions of the harvester head and the log were estimated using off-line machine vision algorihms. Preliminary tests with seven different logs had a maximum error of 12 mm where the mean absolute error between measured and estimated log lengths was 0.09%.