Trajectory Generation for Autonomous Vehicles [chapter]

Vu Trieu Minh
2014 Mechatronics 2013  
Machine learning or statistical artificial intelligence examines averages in data to codify the best approach to make decisions based on the data, and the method notoriously requires large amounts of training data (statistically relevant amounts). On the other hand, data produced by systems that obey physical laws can use those laws as the deterministic structure for artificial intelligence, greatly reducing the requirement for large amounts of data to achieve high-precision performance. This
more » ... ticle examines aspects of deterministic artificial intelligence comprised of self-awareness utilizing auto-trajectory generation followed by change-detection and identification, and then online learning. After the self-awareness algorithm is introduced, several auto-trajectory implementations are compared. Furthermore, learning methods are compared emphasizing recursive least squares and extended least squares learning compared to classical proportional-integral-derivate feedback implementations. Novel learning improvements realized 23.4% decreased mean error, and 34.0% decreased standard deviation compared to the standard recursive least squares optimal estimator error while max error decreased 33.0%. Four autonomous trajectory generation techniques were shown to produce accuracies form 10 -4 to 10 -9 with various computational burdens, providing a menu of implementation options.
doi:10.1007/978-3-319-02294-9_78 fatcat:jv6x7cy22zai5n74neadn2mvz4