A majority of the machining processes in the industry of today are performed using position-controlled machine tools, where conservative feed rates have to be used in order to avoid excessive process forces. Instead of controlling the process forces, the feed rate, and consequently the material removal rate, can be maximized. In turn, this leads to decreased cycle times and cost savings. Furthermore, path planning with respect to time-minimization for milling processes, especially in nonisotropic materials, is not straightforward. This paper presents a model-based adaptive force controller that achieves optimal feed rates, in combination with a learning algorithm to obtain the optimal machining path, in terms of minimizing the milling duration. The proposed solution is evaluated in both simulation and experiments, where an industrial robot is used to perform rough-cut wood milling. Cycle-time reductions of 14% using force control compared to position control were achieved and on average an additional 28% cycle-time reduction with the proposed learning algorithm.