This paper describes an ongoing design and development project of an autonomous patht-racking forest machine. The work is part of a long-term vision in the forest industry of developing an unmanned shuttle that transports timber from the felling area to the main roads for further transportation. The developed prototype system has two modes of operation: Path Learning, in which the human operator drives or remote controls the vehicle along a selected path back and forth from the area of felling to the transportation road. In this phase, position, speed, heading, and the operator’s commands are recorded in the vehicle computer. When the vehicle has been loaded with timber the operator activates Path Tracking mode, which means that the vehicle autonomously drives along the recorded path to the transportation road. A new path-tracking algorithm is introduced, and is demonstrated as superior to standard algorithms, such as Follow the Carrot and Pure Pursuit. This is accomplished by using the recorded data from the path-learning phase. By using the recorded steering angle, the curvature of the path is automatically included in the final steering command. Localization is accomplished by fusing data from Real-Time Kinematic Differential GPS/GLONASS, gyro, wheel odometry, and laser odometry. The laser odometry algorithm works by using consecutive scans to estimate the pose change (position and heading). A search is conducted in pose space to find the optimal fit between the two scans. Test results for path tracking and localization accuracy from runs conducted on the full-sized forest machine are presented.