Umeå University's logo

Smart and autonomous earthmoving equipment enhances energy efficiency,productivity, and safety at construction sites and mines. The innovations provide means to reach high-set sustainability goals and be profitable despite increasing labor shortages. In addition, recent technological breakthroughs in artificial intelligence highlight the potential of superhuman capabilities to further enhance operations. This thesis presents a computational approach to end-to-end optimization of autonomous wheel loaders operating in a dynamic environment. Wheel loaders are mainly used for repeatedly loading material and carrying it to load receivers in quarries and mines. The difficulty lies in that each loading action alters the state of the material pile. The resulting state affects the possible outcomes of the subsequent loading process and, ultimately, the total performance. Thus, the challenge is to achieve both autonomous and high-performance wheel loading over a sequence of tasks. Achieving this requires the ability to predict future outcomes and account for the cumulative effect of loading actions. The thesis constructs a real-time wheel loader simulator, develops world models for sequential loading actions with evolving pile states, formulates the end-to-end optimization problem, and introduces a look-ahead tree search method to solve the problem. These contributions provide insights into utilizing physics-based simulation in combination with machine learning to further improve sustainability in mining and construction.

We investigate how well a physics-based simulator can replicate a real wheel loader performing bucket filling in a pile of soil. The comparison is made using field-test time series of the vehicle motion and actuation forces, loaded mass, and total work. The vehicle was modeled as a rigid multibody system with frictional contacts, driveline, and linear actuators. For the soil, we tested discrete-element models of different resolutions, with and without multiscale acceleration. The spatiotemporal resolution ranged between 50–400 mm and 2–500 ms, and the computational speed was between 1/10,000 to 5 times faster than real time. The simulation-to-reality gap was found to be around 10% and exhibited a weak dependence on the level of fidelity, e.g., compatible with real-time simulation. Furthermore, the sensitivity of an optimized force-feedback controller under transfer between different simulation domains was investigated. The domain bias was observed to cause a performance reduction of 5% despite the domain gap being about 15%.

This paper presents a method for learning world models for wheel loaders performing automatic loading actions on a pile of soil. Data-driven models were learned to output the resulting pile state, loaded mass, time, and work for a single loading cycle given inputs that include a heightmap of the initial pile shape and action parameters for an automatic bucket-filling controller. Long-horizon planning of sequential loading in a dynamically changing environment is thus enabled as repeated model inference. The models, consisting of deep neural networks, were trained on data from a 3D multibody dynamics simulation of over 10,000 random loading actions in gravel piles of different shapes. The accuracy and inference time for predicting the loading performance and the resulting pile state were, on average, 95% in 1.21.2 ms and 97% in 4.54.5 ms, respectively. Long-horizon predictions were found feasible over 40 sequential loading actions.

Simulators are essential for developing autonomous control of off-road vehicles and heavy equipment. They allow automatic testing under safe and controllable conditions, and the generation of large amounts of synthetic and annotated training data necessary for deep learning to be applied [1]. Limiting factors are the computational speed and how accurately the simulator reflects the real system. When the deviation is too large, a controller transfers poorly from the simulated to the real environment. On the other hand, a finely resolved simulator easily becomes too computationally intense and slow for running the necessary number of simulations or keeping realtime pace with hardware in the loop.

We investigate how well a physics-based simulator can be made to match its physical counterpart, a full-scale wheel loader instrumented with motion and force sensors performing a bucket filling operation [2]. The simulated vehicle is represented as a rigid multibody system with nonsmooth contact and driveline dynamics. The terrain model combines descriptions of the frictional-cohesive soil as a continuous solid and particles, discretized in voxels and discrete elements [3]. Strong and stable force coupling with the equipment is mediated via rigid aggregate bodies capturing the bulk mechanics of the soil. The results include analysis of the agreement between a calibrated simulation model and the field tests, and of how the simulation performance and accuracy depend on spatial and temporal resolution. The system’s degrees of freedom range from hundreds to millions and the simulation speed up to ten times faster than realtime. Furthermore, it is investigated how sensitive a deep learning controller is to variations in the simulator environment parameters.

[1]  S. Backman, D. Lindmark, K. Bodin, M. Servin, J. Mörk, and H. Löfgren. Continuous control of an underground loader using deep reinforcement learning. Machines 9(10): 216 (2021).

[2]  K. Aoshima, M. Servin, E. Wadbro. Simulation-Based Optimization of High-Performance Wheel Loading. Proc. 38th Int. Symp. Automation and Robotics in Construction (ISARC), Dubai, UAE (2021).

[3]  M. Servin., T. Berglund., and S. Nystedt. A multiscale model of terrain dynamics for real-time earthmoving simulation. Advanced Modeling and Simulation in Engineering Sciences 8, 11 (2021). 

Having smart and autonomous earthmoving in mind, we explore high-performance wheel loading in a simulated environment. This paper introduces a wheel loader simulator that combines contacting 3D multibody dynamics with a hybrid continuum-particle terrain model, supporting realistic digging forces and soil displacements at real-time performance. A total of 270,000 simulations are run with different loading actions, pile slopes, and soil to analyze how they affect the loading performance. The results suggest that the preferred digging actions should preserve and exploit a steep pile slope. High digging speed favors high productivity, while energy-efficient loading requires a lower dig speed. 

This paper presents a method using data-driven models for selecting actions and predicting the total performance of autonomous wheel loader operations over many loading cycles in a changing environment. The performance includes loaded mass, loading time, work. The data-driven models input the control parameters of a loading action and the heightmap of the initial pile state to output the inference of either the performance or the resulting pile state. By iteratively utilizing the resulting pile state as the initial pile state for consecutive predictions, the prediction method enables long-horizon forecasting. Deep neural networks were trained on data from over 10,000 random loading actions in gravel piles of different shapes using 3D multibody dynamics simulation. The models predict the performance and the resulting pile state with, on average, 95% accuracy in 1.2 ms, and 97% in 4.5 ms, respectively. The performance prediction was found to be even faster in exchange for accuracy by reducing the model size with the lower dimensional representation of the pile state using its slope and curvature. The feasibility of long-horizon predictions was confirmed with 40 sequential loading actions at a large pile. With the aid of a physics-based model, the pile state predictions are kept sufficiently accurate for longer-horizon use.

Wheel loaders in mines and construction sites repeatedly load soil from a pile to load receivers. This task presents a challenging optimization problem since each loading’s performance depends on the pile state, which depends on previous loadings. We investigate an end-toend optimization approach considering future loading outcomes and V-cycle transportation costs. To predict the evolution of the pile state and the loading performance, we use world models that leverage deep neural networks trained on numerous simulated loading cycles. A look-ahead tree search optimizes the sequence of loading actions by evaluating the performance of thousands of action candidates, which expand into subsequent action candidates under the predicted pile states recursively. Test results demonstrate that, over a horizon of 15 sequential loadings, the look-ahead tree search is 6 % more efficient than a greedy strategy, which always selects the action that maximizes the current single loading performance, and 14 % more efficient than using a fixed loading controller optimized for the nominal case.