Abstract: The existing operation path of gravel soil core paving mostly adopts static planning, which fails to consider dynamic changes of the obstacles in a complex operation environment such as bulldozers and dump trucks. To take this case into account, we develop a CA-BINN dynamic path planning model for gravel soil core paving in this paper. Based on the cellular automata (CA) modeling theory, this new model abstracts cellular state information－e.g. dynamic obstacles and paving thickness－from the real-time perceived complex working environment data, and conducts real-time reconstruction of a dynamic construction environment. It inputs this external information into a biological inspired neural network (BINN) algorithm, and reconstructs its neural activity value under the CA framework for calculation of the diversion equation, achieving the dynamic path planning of multi-pile overall paving operation. In this procedure, the paving operation of single material piles takes the indexes－e.g. paving integrity and invalid path ratio－as the objective functions, and dynamic path planning is realized using the three knife CA rule we formulated in the on-site three knife (three times moving material pile) construction. Engineering application shows that our new model not just has high safety and adaptability to complex dynamic operation environments, but can reduce path length, turning times and invalid path ratio by 1.9%, 42.9% and 48% respectively, and improve the paving smoothness by 7% against the static planning model. Compared with manual operation, it increases paving smoothness by 28% and reduces invalid path ratio by 47%, thus improving paving quality and efficiency significantly.

Key words: gravel soil core paving, complex dynamic operation environment, cellular automata, biological inspired neural network (BINN), dynamic path planning