Matlab, Simulink get functions for autonomous driving and deep learning
The Vehicle Dynamics Blockset provides engineers with fully assembled reference application models that simulate driving maneuvers in a 3D environment. Users can visualize ready-made scenes with roads, traffic signs, trees and buildings or adapt these models with their own data.
The blockset provides a standard model architecture that can be used throughout the development process: It contains a library of components for modeling drive, steering, suspension, brakes and other vehicle components. It supports driving and handling analyses and the development of chassis controls.
By integrating vehicle dynamics models into a 3D environment, software for driver assistance systems and automated driving can be tested, for example the behavior of the vehicle during standard driving maneuvers such as a double lane change.
Other products also include new tools or extensions for the development of ADAS systems, e.g. the Automated Driving System Toolbox with the new Driving Scenario Designer App for interactive definition of actors and driving scenarios for testing control and sensor fusion algorithms and the Model Predictive Control Toolbox with ADAS blocks for designing, simulating and implementing adaptive algorithms for speed control and lane keeping.
The Predictive Maintenance Toolbox enables engineers to identify data, design condition indicators and predict and avoid machine failures. Machine data from local files, cloud storage and distributed file systems can be imported for analysis. The toolbox contains reference examples for engines, gearboxes, batteries and other machines, which provide helpful guidelines for the development of your own.
The Neural Network Toolbox now provides a support package to implement deep learning layers and networks designed in TensorFlow Keras. Optimization techniques such as Adam, RMSProp and gradient clipping ensure better training of nets. In addition, networks in the form of directed acyclic graphs (DAG) can be trained faster using several GPUs.
With the GPU Coder, MathWorks provides a tool that automatically converts deep learning algorithms into CUDA code. This allows the algorithms to be executed directly on the GPU. With Release 2018a, networks with topologies in the form of directed acyclic graphs (DAG) and pre-trained networks such as GoogLeNet, ResNet or SegNet are now supported. Another new feature is the generation of C code for deep learning networks on Intel and ARM processors.
Further information: https://uk.mathworks.com