Multisensor Terrain Awareness for Safe and Sustainable Agricultural Robotics in Challenging Environments

Robust terrain awareness is a key enabler for safe, stable, and energy efficient operation of autonomous agricultural robots in unstructured environments. This paper presents a multisensor terrain recognition framework that integrates proprioceptive measurements, including inertial signals, motor currents, and wheel kinematics, with exteroceptive depth-based perception. The proposed approach constructs a physically interpretable feature space capturing both wheel terrain interaction dynamics and local surface geometry. A feature selection strategy based on Minimum Redundancy Maximum Relevance (MRMR) identifies a compact and informative subset of descriptors, significantly reducing dimensionality while preserving discriminative power. Experimental validation on eight representative rigid and deformable terrains demonstrates clear class separability and consistent classification performance using a reduced feature set. The results support scalable multisensor terrain awareness strategies for heterogeneous agricultural robotic platforms operating in challenging field conditions.