Granular flows interacting with rigid bodies are key to machine-terrain interactions in robotics and related fields, and still pose several open problems. Continuum methods and deep learning, separately and in combination, show promise. This research advances machine learning methods for modeling rigid body-driven granular flows, for terrestrial industrial machine and planetary rover (where gravity is an important factor) applications. The original contribution of this research is the application of a subspace machine learning simulation approach for these engineering problems. To generate training datasets, a high-fidelity continuum method, the material point method (MPM) is utilized. Principal component analysis (PCA) is used to reduce the dimensionality of data. It is shown that the first few (8 in our datasets) principal components of our high-dimensional data keep almost the entire variance in data. A graph network simulator (GNS) is trained to learn the underlying subspace dynamics. The learned GNS is then able to predict particle positions and interaction forces with good accuracy, averaging 6×10−5 and 3×10−6 mean squared position error and 7% and 17% mean percentage force error, for excavation and wheel data, respectively. More importantly, PCA significantly enhances the time and memory efficiency of GNS in both training and rollout. This enables GNS to be trained using a single desktop graphics processing unit (GPU) with moderate video random-access memory (VRAM). This also makes the GNS real-time on large-scale 3D physics configurations (700x faster than our continuum method), which is the first demonstration of such significantly accelerated high-fidelity granular flow engineering simulations.

Simulation of wheel-ground and vehicle-ground interactions is very important in many applications. Achieving accuracy and efficiency is challenging for both soft and hard terrains. This is not only because of the simulation and numerical challenges, but also due to the questionable nature of the existing terrain models. For example, the most widely used terramechanics model is not a representative constitutive relation for a full range of dynamic conditions and applications, but rather a parametrization of steady state conditions. In general, the selection and development of the proper constitutive model and the parametrization of the ground properties are very challenging. Here, we present a unified framework for general wheel ground interaction which can be used with different terramechanics models. The framework is based on a complementarity formulation and also uses the concept of kinematic constitutive relations, beside the other known concepts for modelling and parametrizing the soil properties. The framework makes it possible to consider the appropriate modelling of the terrain for a broad range of dynamic behaviours and simulation conditions. We will illustrate the material with several examples for off-road conditions. (c) 2021 ISTVS. Published by Elsevier Ltd. All rights reserved.

This research investigates the development and validation of state-of-the-art high-fidelity m odels o f s oil cutting operations. The accurate and efficient modeling of complex tool-soil interactions is an open problem in the literature. Modeling options that provide more flexibility in trading off accuracy and computational efficiency than current state-ofthe-art continuum or discrete element methods are sought. In this work, two modern numerical methods, the material point method (MPM) and a hybrid approach, are presented with the goal to simulate excavation maneuvers efficiently and with high accuracy. MPM, as an accurate, continuum-based and meshfree method, uses a constitutive model (here, nonlocal granular fluidity model) for computing internal forces to update particle velocities and positions. The hybrid approach, a combination of particle and grid-based methods, avoids explicit integration scheme difficulties and unnecessary computations in the static regime. Visual and quantitative data, including forces on the excavation tool, are collected experimentally to evaluate these two simulation methods with respect to geometry of the soil deformation as well as interaction forces, both as a function of time.

The discrete element method (DEM) is widely seen as one of the more accurate, albeit more computationally demanding approaches for terramechanics modelling. Part of its appeal is its explicit consideration of gravity in the formulation, making it easily applicable to the study of soil in reduced gravity environments. The parallel particles (P-2) approach to terramechanics modelling is an alternate approach to traditional DEM that is computationally more efficient at the cost of some assumptions. Thus far, this method has mostly been applied to soil excavation maneuvers. The goal of this work is to implement and validate the P-2 approach on a single wheel driving over soil in order to evaluate the applicability of the method to the study of wheel-soil interaction. In particular, the work studies how well the method captures the effect of gravity on wheel-soil behaviour. This was done by building a model and first tuning numerical simulation parameters to determine the critical simulation frequency required for stable simulation behaviour and then tuning the physical simulation parameters to obtain physically accurate results. The former were tuned via the convergence of particle settling energy plots for various frequencies. The latter were tuned via comparison to drawbar pull and wheel sinkage data collected from experiments carried out on a single wheel testbed with a martian soil simulant in a reduced gravity environment. Sensitivity of the simulation to model parameters was also analyzed. Simulations produced promising data when compared to experiments as far as predicting experimentally observable trends in drawbar pull and sinkage, but also showed limitations in predicting the exact numerical values of the measured forces. (C) 2020 ISTVS. Published by Elsevier Ltd. All rights reserved.

A new framework is developed for efficient implementation of semi-empirical terramechanics models in multibody dynamics environments. In this approach, for every wheel in contact with soft soil, unilateral contact constraints are added for both the normal direction and the tangent plane. The forces associated with the latter, like traction and rolling resistance, are formulated in this approach as set-valued force laws, their properties being determined by deregularization of the terramechanics relations. As shown in the paper, this leads to the dynamics representation in the form of a linear complementarity problem (LCP). With this formulation, stable simulation of rovers is achieved even at relatively large time steps. In addition, a high-resolution height-field (HF) is employed to model terrain-surface deformation and changes in hardening of soil under the wheel. As a result, the multipass effect is captured in the presented approach. In addition, an extensive set of experiments was conducted using a version of the Juno rover (Juno II). The experimental results are analyzed and compared with the model developed in the paper.

In recent years, realistic simulation of vehicles on soft terrain has increasingly gained importance due to its use in operator training, mission planning and design. In this work we present a method which is tailored to the requirements of real-time simulation of soft soil in Virtual Reality applications. We combine models from terramechanics and soil mechanics in order to represent the interaction of vehicles and their digging tools with a deformable terrain, allowing us to simulate a broad range of vehicles, including earth moving equipment and planetary rovers on soft ground in real-time. We consider the impact of both wheel - ground and tool - ground interactions on the vehicle behavior by creating a fully coupled simulation of the dynamics of the vehicle and its environment. Furthermore we present verification experiments which show that our wheel - ground interaction method yields realistic results.

A new framework is developed in this paper for the efficient implementation of semi-empirical terramechanics models in multibody dynamics environments. In this approach, for every wheel in contact with soft soil, unilateral contact constraints are added to the solver in both the normal direction and the tangent plane. The forces associated with the tangent plane, like traction and rolling resistance, are formulated in this approach as set-valued functions, whose properties are determined by deregularization of the above-mentioned terramechanics relations. As shown in this paper, this leads to the dynamics representation in the form of a linear complementarity problem (LCP). With this formulation, stable simulation of rovers is achieved even with relatively large time steps. In addition, a high-resolution height-field is employed to model terrain-surface deformations and changes in hardening of soil under the wheel. As a result, the multi-pass effect is also captured in our approach. In addition, an extensive set of experiments was conducted using a version of the Juno rover (Juno II). The experimental results are analyzed and compared with the model developed in the paper.

Granular materials exhibit a large number of diverse physical phenomena which makes their numerical simulation challenging. When set in motion they flow almost like a fluid, while they can present high shear strength when at rest. Those macroscopic effects result from the material’s microstructure: a particle skeleton with interlocking particles which stick to and slide across each other, producing soil cohesion and friction. For the purpose of Earthmoving equipment operator training, we developed Parallel Particles (P2), a fast and stable position based granular material simulator which models inter-particle friction and adhesion and captures the physical nature of soil to an extend sufficient for training. Our parallel solver makes the approach scalable and applicable to modern multi-core architectures yielding the simulation speed required in this application. Using a regularization procedure, we successfully model visco-elastic particle interactions on the position level which provides real, physical parameters allowing for intuitive tuning. We employ the proposed technique in an Excavator training simulator and demonstrate that it yields physically plausible results at interactive to real-time simulation rates.

Real-Time Simulation of Mining and Earthmoving Operations: A Level Set-Based Model for Tool-Induced Terrain Deformations D. Holz, A. Azimi, M. Teichmann, S. Mercier Pages 468-477 (2013 Proceedings of the 30th ISARC, Montréal, Canada, ISBN 978-1-62993-294-1, ISSN 2413-5844) Abstract: In this work we present a novel level set-based model for the real-time simulation of soil deformations. A level set defined by a signed distance function and sampled in a regular 3D grid represents and tracks the soil volume under deformation. Moving away from classical 2.5D heightfield representations of soil to a full 3D volume representation allows for improved tracking of cutting tool operations and the simulation of near-vertical or vertical soil faces. The proposed level set representation furthermore provides a versatile mathematical platform for modeling additional effects such as soil slip, and does not suffer from the sampling limitations of commonly used heightfields. Cutting forces applied to the tool are simulated via a formulation based on the Fundamental Equation of Earthmoving, modified to support inclined soil surfaces and transient states of tool motion. Discretisation of this 2D cutting force model with respect to the tool surface allows capturing the effects of irregular 3D terrain shapes on blades and buckets. The surcharge created during cutting operations is tracked in the form of particles and included in the soil failure force computation. We make use of an adaptive, hybrid level set-based and particle-based soil deformation scheme, which allows the soil deformation to be simulated in real-time. Keywords: Physics-based simulation, Computer graphics, Soil mechanics, Deformable surfaces, Virtual reality DOI: https://doi.org/10.22260/ISARC2013/0050 Download fulltext Download BibTex Download Endnote (RIS) TeX Import to Mendeley

Planetary rovers play a key role in modern space exploration. Remotely controlled and operated on unknown terrain, these vehicles run the risk of being damaged, immobilized or simply used in inefficient ways. Their performance depends not only on their design and motion characteristics but also on the performed operation being adequate for the situation and environment they are momentarily in. Consequently, accurate prediction of a specific rover's performance in different situations and environments is a pre-requisite for their efficient and successful deployment. In this work, we combine physical models from terramechanics and soil mechanics to model a rover in highly complex operations, such as excavation tasks with a mounted bulldozer blade. We consider the impact of both wheel-ground and tool-ground interactions on the rover motion system by creating a fully coupled simulation of the dynamics of the rover and its environment. In simulations we show that the rover's mobility and its trajectory are impacted when modifying the way the rover is operated - a direct result of changes in rover-ground interaction.

Realistic physics-based simulation of vehicles on soft ter rain has increasingly gained importance due to their broad application from operator training to mission planni ng and design. In this work we combine models from terramechanics and soil mechanics to represent the interac tion of vehicles and their tools with a deformable terrain, allowing us to simulate a broad range of vehicles, including earth moving equipment and planetary rovers on soft ground in real time. We consider the impact of both wheel–gro und and tool–ground interactions on the vehicle behavior by creating a fully coupled simulation of the dynam ics of the vehicle and its environment. Soil exhibits a multitude of different behaviors depending on material pro perties, soil formations and interactions with the medium, including erosion and compaction. Non-linear stress–stra in relationships as well as highly plastic material deforma tion and flow makes simulation of soil a challenging task. In addit ion, the demands of real-time simulator environments are rather high concerning both realistic visualization an d physical correctness, since for training purposes real li fe scenarios have to be displayed with sufficient accuracy to pr event negative training. Therefore, in Virtual Reality (VR) training simulators for bulldozers, excavators but al so planetary rovers a balance has to be found between interactivity and physical correctness of soil behavior. I n this paper, we present one such solution. In simulation studies we show how the vehicle’s behavior (e.g., its trajec tory and mobility) is impacted when it is operated over areas already modified by its tool or wheels.

The simulation of soil deformation in real-time is a challenging task. Realizing the strengths and weaknesses of particle and mesh-based approaches we propose a hybrid model that combines both. Together with an adaptive sampling method, which effectively reduces the number of particles in the simulation, and a selective update technique our method is applicable in real-time VR environments. Furthermore, in order to account for the high degree of dynamics in soil behavior we consider soil as non-homogeneous and account for its degree of compaction. By incorporating soil mechanical formulations in our model and considering several physically plausible parameters the presented method allows for the simulation of soil as the material empirically investigated by civil engineers and soil mechanicians for decades.

JVRB, 5(2008), no. 7. - The grasping of virtual objects has been an active research field for several years. Solutions providing realistic grasping rely on special hardware or require time-consuming parameterizations. Therefore, we introduce a flexible grasping algorithm enabling grasping without computational complex physics. Objects can be grasped and manipulated with multiple fingers. In addition, multiple objects can be manipulated simultaneously with our approach. Through the usage of contact sensors the technique is easily configurable and versatile enough to be used in different scenarios.