Proprioceptive Terrain Characterization and Adaptive Locomotion on Muddy Surfaces

Locomotion on natural muddy terrains presents significant challenges for ground robots, as the mechanical properties of the substrate can vary significantly over short distances due to changes in water content and particle size distribution. To address this challenge, this paper explores proprioception based sensing methods to characterize mud properties during continuous locomotion and enable real-time robot gait adaptation. We identified three key metrics to represent locomotion-relevant mud properties: penetration resistance, shear strength, and extraction necking force. Our results showed that mud property estimation using the proprioception-based methods matches closely with traditional load cell measurements for all three metrics. This highlights the potential for infering mud properties directly from joint signals. Leveraging the inferred mud properties, we show that a flipper-driven robot could flexibly adjust its appendage insertion depth based on mud strength, effectively preventing locomotion failures across muds of varying consistency. Our results highlighted the value of proprioception-based terrain sensing for guiding robot locomotion adaptation on complex terrains, thus advancing the mobility of robots in diverse and challenging environments.