Abstract: The local path planning methods for autonomous vehicles face the data redundancy and the applicability to the environmental structure in unstructured environments. A modeling method using the sparse representation of 3D Lidar data is presented, which is called potential field dictionary(PFD). The environmental sampling pretreatment results of Lidar signal are sparsely decomposed with Match Pursuit algorithm(MP), whose orthonormal basis is replaced by a preset over-complete DCT dictionary; the sparse decomposition vector is directly applied to the potential field method of the local path planning; dynamic potential field is used to deal with unstructured environments. The results of vehicle test show that the preprocessing result needs little storage space and can express the complexity of the environmental structure better. Afterwards, the PFD algorithm can plan full sequential feasible paths with small sparsity and its performance is superior to the traditional potential field method and RRT* method. The PFD algorithm can reduce the costs of data transmission and storage on the premise of the precision of surrounding information and plan feasible local paths in unstructured environments.

Key words: local path planning, unstructured environments, sparse representation, potential field dictionary