Mobile robots often rely upon systems that render sensor data and perceptual features into costs that can be used in a planner. The behavior that a designer wishes the planner to execute is often clear, while specifying costs that engender this behavior is a much more difficult task. This is particularly apparent when attempting to simultaneously tune many parameters that define the mapping from features to resulting plans. We provide a novel, structured maximum margin approach to learning based on example trajectories demonstrated by a human. The learning problem is transformed into a convex optimization problem and we provide a simple, efficient algorithm that leverages fast planning methods. Finally, we demonstrate the algorithms performance on learning to map features to plans on two different types of input features.