Robotic ground vehicles are commonly used in various road conditions for performing special tasks with different levels of autonomy. This requires the robots to possess a certain degree of perception about the road conditions up ahead and make plans accordingly. It is essential for the robots to be able to quickly adapt their perception for various road conditions they operate. In this paper, an adaptive method for road boundary extraction using 2D LIDAR sensor is presented. A three-stage detection algorithm is utilized for road determination, in which parameter sets are updated adaptively based on a discriminative learning approach. Details of both the learning approach and the detection algorithm are discussed in detail. Experiments were performed on constructed and unconstructed roads to evaluate the performance of the proposed method, and the outcomes were presented in the paper. The results showed that accuracy of road boundary detection increased significantly with the proposed adaptive method. Likewise, significant changes were observed in the required computational times.