Real-Time Nonlinear Model Predictive Control of Heavy-Duty Skid-Steered Mobile Platform for Trajectory Tracking Tasks

TL;DR

The paper addresses real-time trajectory tracking for a heavy-duty skid-steered platform under disturbances. It introduces a multiple-shooting NMPC that fuses SLAM pose data with wheel-velocity measurements and includes dead-zone modeling via a smooth-step function. An NLP is formed by transcribing the OCP on a $SE(2)$-based model with horizon $N=30$ and sampling interval ${\Delta t}=0.1$, employing a quadratic tracking and control-cost and subject to bounds. The approach achieves sub-millisecond, deterministic computation times on-board through warm-starting and bounded iterations, while delivering centimeter-level position accuracy and mm/s-level velocity accuracy across circumference and Lemniscate trajectories. Overall, the method advances real-time NMPC for heavy-duty skid-steered platforms, enabling precise, safe trajectory tracking in challenging environments.

Abstract

This paper presents a framework for real-time optimal controlling of a heavy-duty skid-steered mobile platform for trajectory tracking. The importance of accurate real-time performance of the controller lies in safety considerations of situations where the dynamic system under control is affected by uncertainties and disturbances, and the controller should compensate for such phenomena in order to provide stable performance. A multiple-shooting nonlinear model-predictive control framework is proposed in this paper. This framework benefits from suitable algorithm along with readings from various sensors for genuine real-time performance with extremely high accuracy. The controller is then tested for tracking different trajectories where it demonstrates highly desirable performance in terms of both speed and accuracy. This controller shows remarkable improvement when compared to existing nonlinear model-predictive controllers in the literature that were implemented on skid-steered mobile platforms.

Figures (11)

• Figure 1: Skid-steered mobile platform. This heavy-duty four-wheels robot is hydraulically driven and equipped with a stereo camera system for localization and wheel sensors for retrieving angular velocities.
• Figure 2: Inertial global frame is denoted by $\Sigma_{W}$ and robot's mobile frame by $\Sigma_{b}$. The wheels' reference frames $\Sigma_{w\!F\!L}$, $\Sigma_{w\!F\!R}$, $\Sigma_{w\!R\!R}$ and $\Sigma_{w\!R\!L}$ are placed such that the wheels rotate around their $y$ axes of motion.
• Figure 3: Open-loop test of linear wheel velocities. We command a sine function to the platform and detected marginal performance due to death zones. The measured signal varies only up to reaching a threshold in the reference. The tracking error shows the need of a robust low level controller.
• Figure 4: Experimental setup. The VSLAM system streams the robot pose $\hbox{\boldmath $x$}_{{\scriptsize \hbox{msr}}}$ at 20 Hz to the NMPC block while wheels' angular velocity sensors stream $\,\dot{\!\hbox{\boldmath $\theta$}}_{{\scriptsize \hbox{msr}}}$ at 1 kHz. Our NMPC computes an optimal control $\dot{\!\hbox{\boldmath $\theta$}}^{*}$ every 1 ms to the low-level controller which commands hydraulic motor's RPM $\omega$ runs in Beckoff system. All computation is performed on board.
• Figure 5: From the following experiments, one million samples of the online NMPC are analyzed and showing the percentage of samples faster than certain execution time. These samples include the whole pipeline depicted in Fig. \ref{['fig:overview']} then the NMPC-block execution time can be smaller in reality.