Non-Prehensile Robotic Pushing Strategies for Sloped Terrain and Planetary Surface Conditions

A robot's pushing capability is essential for manipulating and maneuvering objects that are too large to grasp. However, to push an object effectively, it is essential to determine the required force magnitudes and analyze the most efficient pushing strategy. This paper presents an analytical and perception-based learning framework of non-prehensile pushing on inclined terrain for planetary robotics. The study investigates the robot’s ability to move objects effectively through non-prehensile pushing by combining analytical modeling, dynamic simulation, and experimental validation on uneven slopes. A quasi-static analytical framework is developed to model the pushing dynamics of each robot under different conditions. Building on this foundation, a neural network model is trained to predict surface inclinations from depth images. A physics-based MuJoCo environment is used to generate depth–normal training data and test pushing behavior on uneven heightfields, similar to lunar surface. Additionally, real-robot experiments using a Z1 manipulator with MoveIt demonstrate in real world setup of the proposed system. These findings provide insights that support the design of robotic systems for planetary surface missions, where pushing may be necessary to move rocks, manipulate samples, or assist with infrastructure tasks.