Design and experimental evaluation of model predictive control for autonomous articulated dump trucks

Articulated dump trucks (ADTs) serve as key vehicles in mining operations, as well as in urban construction sites. As demand for automation in these industries increases, using model predictive controllers (MPCs) for the autonomous operation of these vehicles in the context of trajectory-following controllers is becoming popular due to their ability to predict and optimize vehicle performance in real time. However, successful implementation of these controllers requires accurate models of the vehicle. Previous research has only explored MPC concepts using well-established kinematic bicycle models. In order to improve the performance of the existing MPCs, this paper presents dedicated models for different operation scenarios (forward and backwards driving) and different ADT sizes (full-sized and compact). These models capture and take advantage of the scenario- and size-related specifics, resulting in better backwards driving performance across all ADT sizes and also enhancements in forward driving, specifically in small compact ADTs. The proposed controllers are tested in unpaved terrain of a mining field on a 1.3-t compact ADT and a 20-t full-sized ADT. Experimental results indicate that the controllers successfully fulfill the requirements and improve tracking compared to state-of-the-art methods by 70 percent in terms of mean absolute lateral error (MAE) when a comparison is possible. Background An MPC is a controller that uses the dynamic model of the system to predict its future behavior and therefore optimize the control input over a finite time horizon. Besides being able to consider the future dynamic behavior of the vehicle, these controllers offer useful features: specifically, the ability to consider dead time, which is also known as transport delay and is the time delay between applying an input and observing its effect. This feature makes MPCs a viable option to use in ADTs, as these vehicles have inherent dead times that cannot be neglected. Fig. 1 Results for the full-sized ADT. Method As MPCs require system models for optimization, the focus of this paper is to establish suitable vehicle models. Starting from the well-established kinematic bicycle model of an ADT, the model is first adjusted to be able to embody the considerable actuator delays. These delays, which are up to 0.5 s in the acceleration, deceleration and articulation in the case of the studied vehicles, play an important role in the control. Neglecting these delays, which is to falsely assume that the vehicle can instantly realize acceleration, deceleration or articulation commands, leads to snaking behavior around the given trajectory or to instable controllers at higher speeds. Snaking behavior is the oscillatory side-to-side motion of the vehicle around the reference trajectory as it attempts to stay on the trajectory. After the establishment of the baseline model with the delays, the model is adjusted for driving forward and backwards. Furthermore, to establish a model that better mirrors the vehicle behavior also in terms of the control input, a secondary set of models is established for ADTs with a different steering option. The background to this improvement is that compact ADTs (c-ADTs) might have only “articulation angle” as a possible control input, while full-sized ADTs (f-ADTs) and the models established so far in the literature all assume “articulation angle rate” as the input. Establishing scenario-based (forward and backwards driving) and actuator-type-specific models of the ADTs (with articulation or articulation angle rate as input) gives the MPC a more realistic model of the vehicle, leading to better compatibility between the controller and the vehicles. The well-established and already existing model is a front-axle-referencing model for an f-ADT (FARf). The new models are the rear-axle-referencing model for driving backwards in the f-ADT (RARf) and the same models for the c-ADTs (FARc and FARc). So far in the literature, an FARf model was used for all scenarios and sizes. Fig. 2 Results for the compact ADT. The importance of applying accurate models shows itself more when driving at higher speeds. In other words, experiments show that when driving at lower speeds (especially under 3 m/s) or on relatively straight roads, neglecting the delays does not cause a noticeable lateral error or snaking effect. Therefore, incorporating a system model and delays offers the potential to drive at higher speeds. Results and discussion The controllers were tested in experiments on real consumer vehicles, a Bell B30E and a Wacker Neuson 1501, representing the two major ADT sizes and the two mentioned control input types. Figure 1 shows the results for the f-ADT, and Fig. 2 shows the results for only driving backwards in the c-ADT. Table 1 also summarizes the results for the f-ADT and c-ADT, respectively. The results show a decrease in the lateral error when driving at higher velocities compared to the state-of-the-art methods when a comparison is possible. Specifically, using tailored models made the following improvements: Driving backwards with the f-ADT became possible as an MPC with the existing model became infeasible. The mean absolute lateral error (MAE) while driving backwards was 15.9 cm. In driving backwards with the c-ADT, the MAE was 11.5 cm compared to 40.1 cm when using the existing model. The maximum speed reached by the f-ADT was 5.2 m/s, which is higher than the usual speed in the current literature. Conclusion In this work, models and MPCs were developed and implemented for autonomous forward and backwards driving for both compact ADTs with faster steering dynamics and full-sized ADTs. By moving the reference point of the model to the rear axle, this study managed to address the issue with the nonminimum phase behavior of the front-axle referencing when driving backwards and achieved sufficient accuracy for driving backwards in narrow passages, such as crushers. Furthermore, by using system identification techniques, this study was able to directly use articulation angle as the control variable in the model of compact ADTs and to benefit from their faster steering dynamics. The experimental validation of the MPCs shows that the MPCs met the specified requirements and show potential to be used in real-world applications. In future research, the trajectory planning for ADTs will be studied in more detail, and model-based algorithms will be utilized. Furthermore, different methods for reducing or estimating model uncertainties will be studied and evaluated. Additionally, methods for online estimation of the model parameters will be studied in detail to enable real-time adaptation of the model parameters to changes in the vehicle dynamics or road conditions. References A list of all references is available in the full paper.

Design and experimental evaluation of model predictive control for autonomous articulated dump trucks

Authored by: Arash Shahirpour and Tim Reuscher

To view this article you must be a member of SME/UCA.

Caterpillar Announces Leadership Transition in Resource Industries

Copper squeeze breathes new life into old mines

Cartier-owner Richemont weighs joining De Beers acquisition race, source says

Caterpillar Announces Leadership Transition in Resource Industries

Copper squeeze breathes new life into old mines

Cartier-owner Richemont weighs joining De Beers acquisition race, source says

To view this article you must be a member of SME/UCA.

Using overall equipment effectiveness as a driver for improving the productivity of continuous mining systems

Drill and blast decision support to optimize hole position, explosives and drill bit clusters

Related Articles

Rare earth elements in the global economy: Usage, recovery and the quest for supply security — A review

A systematic evaluation of localization signals in the advancing openpit mine Aitik

From foundation to future: Revisiting AI integration in mining engineering education through current perspectives of students, educators and industry

Iron ore tailings reprocessing by electrostatic separation plus two-stage froth flotation

Caterpillar Announces Leadership Transition in Resource Industries

Copper squeeze breathes new life into old mines

Cartier-owner Richemont weighs joining De Beers acquisition race, source says

Caterpillar Announces Leadership Transition in Resource Industries

Copper squeeze breathes new life into old mines

Cartier-owner Richemont weighs joining De Beers acquisition race, source says

Use Single Sign On