CNN Heuristic Maps for Faster Path Search in Complex Environments

The patent applies preliminary action by pre-learning heuristic functions using convolutional neural networks before actual path searching. The system learns heuristic functions from training data that represent environmental features and task requirements, storing this knowledge in advance. During actual path searching, the pre-learned heuristic functions are directly applied, eliminating the need for time-consuming real-time learning and significantly reducing search time while maintaining high accuracy in complex environments.

The patent implements universality by designing a convolutional neural network that learns universal heuristic functions applicable to multiple different environments and tasks. The CNN architecture processes various types of input data (environmental maps, task specifications) and generates heuristic functions that can be applied across different path planning scenarios. This multi-functional approach allows a single learning system to handle diverse applications, from navigation in different environments to various task types, thereby improving versatility while managing system complexity through a unified framework.