Abstract:Legged locomotion of robots has advantages in reducing payload in contexts such as travel over deserts or in planet surfaces. A recent study (Li et al. 2013) partially addresses this issue by examining legged locomotion over granular media (GM). However, they miss one extremely significant fact. When the robot’s wheels (legs) run over GM, the granules are set into motion. Hence, unlike the study of Li et al. (2013), the viscosity of the GM must be included to simulate the kinematic energy loss in striking and passing through the GM. Here the locomotion in their experiments is re-examined using an advanced Navier-Stokes framework with a parameterized granular viscosity. It is found that the performance efficiency of a robot, measured by the maximum speed attainable, follows a six-parameter sigmoid curve when plotted against rotating frequency. A correct scaling for the turning point of the sigmoid curve involves the footprint size, rotation frequency and weight of the robot. Our proposed granular response to a load, or the ‘influencing domain’ concept points out that there is no hydrostatic balance within granular material. The balance is a synergic action of multi-body solids. A solid (of whatever density) may stay in equilibrium at an arbitrary depth inside the GM. It is shown that there exists only a minimum set-in depth and there is no maximum or optimal depth. The set-in depth of a moving robot is a combination of its weight, footprint, thrusting/stroking frequency, surface property of the legs against GM with which it has direct contact, and internal mechanical properties of the GM. If the vehicle’s working environment is known, the wheel-granular interaction and the granular mechanical properties can be grouped together. The unitless combination of the other three can form invariants to scale the performance of various designs of wheels/legs. Wider wheel/leg widths increase the maximum achievable speed if all other parameters are unchanged.

Optimal Control of Granular Material

A Thermodynamics-Informed Neural Network for Elastoplastic Constitutive Modeling of Granular Materials

Granular Gym: High Performance Simulation for Robotic Tasks with Granular Materials

Optimal Parametric Control of Nonlinear Random Vibrating Systems

Developing Physics-Informed Neural Networks for Model Predictive Control of Periodic Counter-Current Chromatography

AutomaChef: A Physics-informed Demonstration-guided Learning Framework for Granular Material Manipulation

Neural Field Dynamics Model for Granular Object Piles Manipulation

Subspace Graph Physics: Real-Time Rigid Body-Driven Granular Flow Simulation

Inferring the Material Properties of Granular Media for Robotic Tasks

Learning Robotic Manipulation of Granular Media

Subspace graph networks for real-time granular flow simulation with applications to machine-terrain interactions

Gaussian Splatting Visual MPC for Granular Media Manipulation

Graph Neural Network-based surrogate model for granular flows

An efficient algorithm for granular dynamics simulation with complex-shaped objects

Complex Network Analysis of Force Chain Structure for Granular Materials under Loading and Unloading Conditions

Programming Active Cohesive Granular Matter with Mechanically Induced Phase Changes

Machine Learning Aided Modeling of Granular Materials: A Review

Online Reinforcement Learning Neural Network Controller Design for Nanomanipulation

Stress fields in granular material and implications for performance of robot locomotion over granular media

Hybrid model development and nonlinear model predictive control implementation for continuous dry granulation process