For the safe path planning of mobile robots in unknown environments, the paper proposes a locally linked Hopfield artificial neural network （ANN） pl anner. For the environments of arbitrary shape, without the learning process, ANN plans a safe path with the consideration of both“too close”and“too   far”. For the effective application on a single processor to plan a path on-line, the simulation based on constrained distance transformation is propo  osed to accelerate the propagation of the numerical potential field of ANN. Simulations demonstrate the method has high real-time ability and adaptabili ty to environments.