Development of autonomous farm vehicle navigation

The state of autonomous field- and service robotics has developed slowly but steadily since the 80's. Attempts to create an autonomous robot for farm conditions have been the goal for many research and product development projects. This goal has remained mostly unreachable due to the complexity of accurate navigation techniques. Recent development in Simultaneous Localization And Mapping (SLAM) technique and laser range scanner based dead reckoning has provided encouraging results for mobile robot navigation. The purpose of this thesis was to survey these techniques and develop autonomous navigation and route repetition for a farm vehicle to a small husbandry farms. In this thesis the current state of SLAM techniques was surveyed and introduced as well as some closely related techniques like Kalman filter. The robot pose estimation was done by developing a kinematic model for a 4 wheel steering vehicle and by implementing Iterative Dual Correspondence (IDC) algorithm for scan alignment. The IDC-algorithms performance was tested for both time and accuracy. The results indicated that the accuracy is similar to what has been presented in the literature. The developed kinematic model was also tested through simulation and by implementing it to a prototype. Also route recording and reconstruction techniques were researched and implemented to generate a route trajectory for the vehicle to follow. The results indicate that the implemented methods worked partially. The kinematic model worked nicely as well as the IDC-algorithm in accuracy. Considering the execution speed of the algorithm the results are not so flattering. The processing time takes too long for successful real-time implementation in the target prototype. Route was recorded by saving robot pose at constant distance intervals. Route was reconstructed using piecevise Bezier curve interpolation. The implemented method provided good means to reconstruct a continuous planar route trajectory except for the curve length calculation. The route repetition was done using direct control without feedback by measuring the distance traveled and taking the curvature of the reconstructed route as a function of distance. The simulated route repetition method failed as well as the real tests. It is advisable to generate a good feedback control strategy for the route tracking