Automated Driving: Trajectory Planning for Unmanned Trucks in Mining Areas

Basic Background

Open-pit mining presents a typical application scenario for automated driving. Contrasting the navigation of small passenger cars on structured urban roads, the movement of mining trucks in unstructured mining environments, particularly during loading and unloading operations, is significantly more intricate. Achieving safe, efficient, and high-quality trajectory planning in these areas is crucial for enhancing the efficiency of unmanned mining operations.

Development Challenges

The complexities of trajectory planning in loading and unloading areas in open-pit mining scenarios are evident in several aspects:

Complex Kinematics of Mining Vehicles:

Mining vehicles can maneuver forward and backward multiple times, resulting in non-smooth path trajectories. Due to the challenges in tracking control during reverse driving, the curvature and speed of reverse driving paths are often set more strictly. Each segment of forward/backward travel should not be excessively short.

Complex Environmental Constraints on Mining Vehicles:

These vehicles must avoid collisions with randomly placed obstacles in the environment. Additionally, they must adhere to additional spatial constraints, such as safety construction regulations requiring the vehicle to reach a specific angle while reversing under the shovel of the loading machine.

Complex Driving Habits of Mining Vehicles:

Apart from the mentioned rigid constraints, the driving behavior of mining vehicles should closely resemble that of human drivers, aiming for anthropomorphism. For instance, the reverse driving path should not be excessively long, and the number of vehicle maneuvers should be minimized.

The complexity outlined above makes trajectory planning in mining loading and unloading scenarios a problem involving logical expression-based optimal control. Numerically solving this problem involves dealing with mixed-integer nonlinear programming propositions containing conditional constraints, for which there are currently no dedicated solving tools.

Technological Approach

To address these challenges, Professor Li Bai's team from the School of Mechanical and Vehicle Engineering at Hunan University proposed a hierarchical planning strategy. At the upper planning level, they introduced an optimization-enhanced hybrid A* search algorithm. This algorithm considers certain complex factors involving logical expression-based constraints, obtaining globally optimal approximate feasible solutions. The lower-level planning inherits the approximate feasible solutions outputted by the upper level. Based on these solutions, it simplifies the cost functions and constraints of the original proposition, constructing smaller-scale nonlinear programming propositions for quick resolution. To ensure that the simplification in the lower level does not lead to results contradicting the complex factors considered sufficiently by the upper level, they added constraints in the nonlinear programming proposition to ensure ordered transmission between levels.

Development Experience

The algorithm's development spanned nearly two years and underwent over ten rounds of iterative improvements, integrating actual operational requirements in mining areas and algorithmic performance. To enhance development efficiency, the research team conducted numerous indoor small-scale experiments, extensively validating the algorithm's prototype. Infrared sensors for motion capture were utilized for positioning indoor miniature vehicles, achieving sub-millimeter-level accuracy. Following engineering refinement and adaptation, the aforementioned trajectory planning algorithm was deployed on automated driving mining trucks in the Smart Mining Project of Sijiang Tailings, Nanjing. After comparison measurement certification by the China National Institute of Metrology, the project's unmanned pure electric mining trucks exceeded manned fuel-powered mining trucks comprehensively in production efficiency, safety, and economics. This project stands as the first third-party-certified unmanned driving project surpassing manual efficiency.