The global construction industry faces the dual challenges of enhancing productivity and ensuring safety, with the automation and unmanned operation of construction equipment emerging as key solutions. However, conventional autonomous driving technolo...
The global construction industry faces the dual challenges of enhancing productivity and ensuring safety, with the automation and unmanned operation of construction equipment emerging as key solutions. However, conventional autonomous driving technologies have shown clear limitations in effectively addressing the complex, variable environments of construction sites and the unique characteristics of construction machinery. To overcome these technical challenges, this study proposes the Model Predictive Path Integral (MPPI) control method.
This paper aims to prevent rollover accidents that may occur during off-road driving of articulated construction vehicles by developing an MPPI-based autonomous driving algorithm. The proposed approach overcomes the limitations of traditional path planning methods by precisely reflecting the vehicle’s dynamic characteristics and constraints, with a particular focus on preventing rollovers that can lead to serious accidents on construction sites.
To this end, an MPPI framework optimized for articulated vehicles was designed to capture the
vehicle’s dynamic behavior and integrate real-time rollover prevention and active obstacle avoidance. To accurately represent the unstructured terrain of construction sites, a grid map containing normal vector information was generated from 3D point cloud data. The RANSAC (Random Sample Consensus) algorithm was employed to robustly estimate normal vectors by minimizing the influence of noise and outliers in the terrain data. RANSAC iteratively samples random subsets to generate models and selects the model supported by the largest number of inliers, making it highly effective for extracting reliable parameters from real-world data prone to measurement errors. Based on the constructed grid map and the vehicle’s current yaw angle, a real-time roll angle estimation algorithm was developed.
The MPPI controller probabilistically samples multiple candidate control sequences and predicts future states for each sequence using a nonlinear dynamic model. The optimal control input is then determined by evaluating a cost function that comprehensively considers rollover risk, collision risk, path tracking accuracy, and terrain adaptability. For rollover prevention, the controller calculates dynamic stability indices in real time using the vehicle’s dynamic state and estimated terrain information, integrating these indices as key elements in the MPPI cost function. If the calculated stability index or estimated roll angle exceeds a predefined safety threshold, a large penalty is imposed on the corresponding control sequence, effectively excluding unsafe trajectories through strong constraints. Furthermore, when encountering unexpected obstacles, the MPPI immediately incorporates obstacle information into the cost function, enabling rapid replanning of safe avoidance paths.
The performance of the proposed integrated controller in rollover prevention and obstacle avoidance was comprehensively validated in various rough-terrain driving scenarios using Vortex Studio, a simulation environment specialized for construction equipment. The evaluation focused on the appropriateness of path generation, path tracking accuracy, safety during operation, and real-time control feasibility