MATLAB 2018a has just been released and contains a host of new features to both the MATLAB core program as well as its Simulink platform.
First up is the new Vehicle Dynamics Blockset update to Simulink.
Users are now able to view their simulations in a 3D environment thanks to the Unreal Engine, which allows them to better visualize and understand the vehicle’s dynamic responses. The blockset comes with several prebuilt scenes that enable you to try out your different scenarios. Additional scenes can be obtained through the Vehicle Dynamics Blockset interface for Unreal Engine 4 support package. The support package also includes project files you can use in the Unreal Engine editors to customize scenes for your own needs. And, naturally, you can add your own custom blocksets to the predefined system and see how they affect overall system performance.
The blockset also contains a number of driving maneuvers ranging from double lane change to swept sine waves, enabling users to test their systems under different driving conditions.
Additionally, the blockset contains chassis control parameters, which allow users to design closed-loop controllers such as ABS or yaw control.
And, of course, being 2018 you can bet your bottom dollar that there are some automated vehicle features in the new release. Indeed, the Automated Driving Toolbox provides exactly that, allowing users to develop lane changing algorithms, automated emergency braking and hazard avoidance.
In terms of the core program, many of the upgrades are focussed on deep learning. For the benefit of the uninitiated, deep learning is a subset of machine learning that is inspired by thought processes in the human brain (specifically, deep learning programs attempt to copy the activity of layers of neurons in the neocortex).
In other words, it learns good, and is particularly useful for recognizing patterns in analog inputs such as images or music.
New deep learning features include:
- Regression and bidirectional LSTMs for continuous, time series outputs
- Automatic validation of custom layers to check for data size and type consistency
- Additional optimizers for training: ADAM and RMSprop
- Train Directed Acyclic Graph (DAG) networks in parallel and on multiple GPUs
- Support for DAG networks, including GoogLeNet, ResNet-50, ResNet-101, Inception-v3 and SegNet
- Support for Intel and ARM processors
- The ability to generate CUDA code that integrates with TensorRT
So, there you have it. MATLAB 2018a is smarter and prettier than previous releases.
It’s on sale right now, and you can take a look at the release notes over at this link.