Machine learning for autonomous vehicles

To perform complex subjects in perception, decision making and motion planning, driverless vehicles depend on machine learning. However, automotive software safety standards have not completely matured to handle machine learning safety concerns such as interpretability, verification, and performance limits. In this paper, we review recent developments in planning, decision-making, and perception for autonomous cars that resulted in significant functional gains, with numerous prototypes currently operating on our streets and roads. However, issues endure in ensuring assured performance and dependability in all driving conditions. Many businesses are evolving self-driving automobiles, which are likely to be widely used in the near future. Considering this, there is no consensus on effective techniques for testing, debugging, and certifying the system performance. One key challenge is that many unmanned driving systems use ML (machine learning) elements like DNN (deep neural networks), which seem to be hard to interpret in formal terms. It is necessary to develop planning solutions that provide safe and scheme performance in complex, crowded situations, such as predicting the unanticipated interaction with other traffic participants.