Multilayer feedforward networks are universal approximators
Neural Networks
Learning internal representations by error propagation
Parallel distributed processing: explorations in the microstructure of cognition, vol. 1
Intelligent transportation systems-Enabling technologies
Mathematical and Computer Modelling: An International Journal
| Hi-index | 0.00 |
A solution to autonomous lateral vehicle guidance using a neurocontroller that can learn from measured human-driving data without knowledge of the physical car parameters is discussed. Simulations and practical tests confirm that a small-size feedforward autonomous neural network (21 neurons) can learn to steer a vehicle at high speeds only from looking at human-driving examples. In this way, the network learns the total closed-loop behavior, including the nonlinear dynamics of the vehicle and the driver's individual driving style. The main result of practical investigations is that the neutral controller trained on human-driving examples exhibits an aperiodic behavior that does not vanish at higher speeds (tests performed up to 130 km/h) and produces fewer lateral deviations than the linear state controller.