The initial phase of the project will focus on the identification and adaptation of suitable pseudo-soft or reconfigurable mobility systems (e.g., ARC’s SUPERball v2, JPL’s PUFFER). In the second phase, we will define a sensor fusion (e.g. UKF) and machine learning algorithm (e.g. MPC, RL) able to characterize terrain features (e.g. slope, substrate, barriers) and evaluate robot performance (e.g. speed, slip, stability) using distributed proprioceptive sensors. As a final step, our terrain detection algorithm will be integrated with existing locomotion control policies and JPL’s TARMAC (Terrain Adaptive Reconfiguration of Mobility by Automatic Control) path-planning algorithm to enable autonomous crossing of adverse terrains.