{"id":3432,"date":"2019-09-25T12:00:02","date_gmt":"2019-09-25T10:00:02","guid":{"rendered":"https:\/\/blogs.mathworks.com\/student-lounge\/?p=3432"},"modified":"2021-12-08T13:39:34","modified_gmt":"2021-12-08T18:39:34","slug":"robotics-r2019b","status":"publish","type":"post","link":"https:\/\/blogs.mathworks.com\/student-lounge\/2019\/09\/25\/robotics-r2019b\/","title":{"rendered":"What’s New in MATLAB R2019b for Robotics?"},"content":{"rendered":"

In MATLAB R2019b, there are major product updates and new toolboxes. You can look at the release notes<\/a> for a complete list, but this blog has the capabilities I find most exciting and relevant to robotics and autonomous systems.<\/p>\n

Robot Simulation <\/strong><\/h1>\n

Robotics System Toolbox<\/a> retains a collection of specialized modeling and simulation tools for different types of robots. It specifically focuses on providing low-fidelity robot models, where the low-level controls are representative of a realistic system, so you can quickly test higher-level algorithms for motion planning, task planning, and behavior. In addition, there are more detailed analysis tools that provide support for low-level control design workflows.<\/p>\n

Manipulators<\/h2>\n

There is now a library of commercial robot models<\/a>, low-fidelity joint-space<\/a> and task-space<\/a> motion models that capture the behavior of common closed-loop control strategies, and collision detection<\/a> between simple shapes or meshes. All of these integrate with the tools for kinematics and dynamics analysis and trajectory planning that already existed in previous releases of Robotics System Toolbox.<\/p>\n

Make sure to check out the new robot manipulator examples that range from basic trajectory planning<\/a> to complete pick-and-place task planning<\/a>.<\/p>\n

\"\"<\/p>\n

Ground Vehicles\/Mobile Robots<\/h2>\n

For mobile robots, there are new low-fidelity kinematic motion models<\/a> for different types of mobile platforms \u2013 most commonly differential drive and car-like vehicles. These models can connect with the sensor models and path planners included in Navigation Toolbox (see the next section for more detail) to prototype autonomous navigation algorithms. They can also be used to design model-based path planners and followers such as Model Predictive Control<\/a>.<\/p>\n

Make sure to check out the new warehouse robot examples that range from basic path planning for a single robot<\/a> to task planning and coordination for a swarm of robots<\/a>.<\/p>\n

\"\"<\/p>\n

Gazebo Cosimulation<\/h2>\n

There is now a direct interface between Simulink and the Gazebo simulator<\/a>.<\/p>\n

We have been able to connect to Gazebo through ROS in the past, but with a direct interface a Simulink model can now drive execution of the Gazebo world so the dynamics in both tools are always in sync. This will enable you to use Simulink for designing control algorithms<\/a> while getting synthetic sensor data<\/a> (such as images and lidar) from Gazebo.<\/p>\n

My personal favorite application of this technology is towards the training of reinforcement learning (RL) agents, which typically require the action at one time step to lead to an observation at the next time step. Below is my attempt at training a simulated Kinova Gen3 robot arm<\/a> to place its gripper near an object using the Deep Deterministic Policy Gradient (DDPG) algorithm<\/a>. With Gazebo as my environment, the observation input includes images from the arm-mounted camera so I can train agents based on convolutional neural networks (CNNs)<\/a>.<\/p>\n

\"\"<\/p>\n

Learn more about Robotics System Toolbox below.<\/p>\n