A Coupled DEM--MBD Framework for High-Fidelity Simulation of Multi-legged Robot Locomotion on Granular Terrain
This paper presents a high-fidelity, GPU-accelerated coupled DEM-MBD framework that accurately simulates the mutual interaction between a twelve-legged spherical robot and deformable granular terrain by resolving complex contact mechanics and terrain evolution during locomotion.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine a robot trying to walk across a beach. On solid ground, its feet simply push against a surface that does not move. But on sand, every step changes the ground beneath it. The sand shifts, piles up, and sinks, creating a new shape that dictates where the robot can go next. This is the fundamental challenge of moving across loose, granular terrain like the sand found on the Moon or Mars. For decades, engineers have struggled to predict how a machine will behave in such an environment. Traditional methods often treat the ground as a static, unchanging surface or rely on simplified rules that cannot capture the chaotic, shifting nature of individual grains. To design robots that can truly explore these alien worlds, scientists need a way to simulate the complex, two-way conversation between a machine's legs and the millions of tiny particles that make up the ground.
A team of researchers at Beihang University and Sun Yat-sen University has developed a new computer framework to solve this problem. They created a high-fidelity simulation that couples two different ways of modeling the world: one that treats the robot as a system of connected, moving parts, and another that treats the ground as a collection of millions of individual, interacting spheres. Instead of assuming the ground is rigid or follows a pre-set pattern, their model lets the terrain evolve in real time. As the robot moves, its feet push into the sand, causing particles to rearrange, roll, and pile up. In response, the changing shape of the ground alters the forces acting on the robot, which in turn changes how the robot moves. This continuous loop allows the researchers to watch a virtual robot navigate a granular landscape with a level of detail previously impossible to achieve.
The robot at the center of this study is a unique design: a twelve-legged spherical machine with a radial skeleton. Unlike a rover with wheels, this robot moves by extending and retracting its legs in a coordinated sequence, effectively rolling itself forward. On a hard floor, this motion is predictable. On loose sand, however, the outcome is uncertain. The researchers wanted to see if their new simulation could capture the messy reality of this interaction. They built a digital environment filled with over 200,000 virtual sand particles, each representing a grain of regolith. The robot's feet were modeled as triangular surfaces that could press into this sea of particles. The simulation tracked every collision, every force, and every moment of rotation, calculating how the robot's weight caused the sand to sink and how the sand's resistance pushed back against the robot's legs.
The results of these simulations revealed a dynamic and often surprising picture of locomotion. When the robot attempted to move in a circle under Earth's gravity, the simulation showed that it did not rely on all twelve legs at once. Instead, only a small number of legs—typically four—carried the load at any given moment. The other legs hovered or touched the ground without bearing weight. As the robot rolled, the active legs shifted, and the load transferred from one set of limbs to another in a fluid, emergent pattern. The ground itself reacted visibly. The simulation showed the sand sinking up to 26.51 millimeters in some spots while piling up in others, creating a distinct, irregular footprint that matched the robot's path. The robot did not just move over the sand; it reshaped the sand, and the reshaped sand dictated the robot's next move.
The researchers also tested the robot under simulated lunar gravity, which is much weaker than Earth's. In this low-gravity environment, the robot moved in an S-shaped pattern. The simulation showed that the reduced gravity led to less sinking and a smaller disturbance of the sand, with the maximum depth of penetration dropping to just over one millimeter. Despite the different conditions, the core behavior remained the same: the robot's motion was not a simple, pre-programmed path but a negotiation with the terrain. The simulation captured how the robot's legs slipped, how the sand flowed around the feet, and how the machine adjusted its balance in response to these forces. The model successfully predicted the robot's trajectory, the forces on its legs, and the final state of the disturbed ground, matching the physics of the interaction with high precision.
To ensure their new tool was accurate, the researchers first tested it against known physical phenomena. They simulated a simple collision between two balls and compared the results to real-world experiments, finding that their model correctly predicted how the balls bounced and transferred energy. They also simulated sand pouring out of a hopper, a classic test of granular flow, and found that their digital sand behaved almost exactly like real sand, flowing at the same rate and forming the same patterns. Finally, they dropped a heavy cube onto a bed of sand and compared their results to a different, well-established simulation software. The two programs agreed closely on how the cube bounced, how it settled, and how the sand pushed back, giving the researchers confidence that their new framework was reliable.
The significance of this work lies in its ability to reveal details that simpler models miss. Previous methods often assumed the ground was a fixed boundary or used rough approximations of how soil behaves. This new framework shows that the support a robot receives is not a fixed condition but a constantly changing result of the interaction between the machine and the environment. It demonstrates that a robot's path is not just determined by its motors and legs, but by the chaotic, shifting support of the ground beneath it. By capturing these details, the researchers have provided a powerful tool for designing future explorers. Before sending a robot to the Moon or Mars, engineers can now run these high-fidelity simulations to see how a specific design will handle the loose, shifting soil of another world, ensuring that the machine can navigate the terrain it will actually encounter.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.