Learning multiobjective rough terrain traversability

We present a method that uses high-resolution topography data of rough terrain, and ground vehicle simulation, to predict traversability. Traversability is expressed as three independent measures: the ability traverse terrain at target speed, energy consumption, acceleration. The measures are continuous reflect different objectives for planning go beyond binary classification. A deep neural network trained traversability from local heightmap speed. To produce training data, we use an articulated with wheeled bogie suspensions procedurally generated terrains. evaluate model on laser-scanned forest terrains, previously unseen by model. predicts accuracy 90%. Predictions rely features high-dimensional surpass roughness slope relative heading. Correlations show complementary each other. With inference speed 3000 times faster than truth simulation trivially parallelizable, well suited analysis optimal path over large areas.