{"id":4431,"date":"2020-05-19T13:27:23","date_gmt":"2020-05-19T17:27:23","guid":{"rendered":"https:\/\/blogs.mathworks.com\/deep-learning\/?p=4431"},"modified":"2021-04-06T15:48:24","modified_gmt":"2021-04-06T19:48:24","slug":"autonomous-systems-mathworks-simulink-microsoft-project-bonsai","status":"publish","type":"post","link":"https:\/\/blogs.mathworks.com\/deep-learning\/2020\/05\/19\/autonomous-systems-mathworks-simulink-microsoft-project-bonsai\/","title":{"rendered":"Building Autonomous Systems with Simulink and Microsoft\u2019s Project Bonsai"},"content":{"rendered":"Below is a guest post from Aditya Baru, product marketing manager for our AI product group. \u00a0He will be talking about a new partnership between MathWorks and Microsoft.<\/em>\r\n
<\/h6>\r\nI\u2019m glad to announce that MathWorks is partnering with Microsoft to help engineers develop autonomous systems for industrial and manufacturing applications.\r\n
<\/h6>\r\nThis partnership focuses on the integration between Simulink<\/strong>, our platform for simulation and Model-Based Design and Project Bonsai<\/strong>, Microsoft\u2019s cloud-based platform for designing autonomous systems through a combination of machine teaching and reinforcement learning techniques.\r\n
<\/h6>\r\nProject Bonsai enables engineers and domain experts to use machine teaching techniques to break down complex problems into smaller parts that can be solved faster using AI algorithms. These algorithms learn to solve individual problems in a trial-and-error manner by trying out a wide range of possible actions to take in a given scenario and then selecting the best one.\r\n
<\/h6>\r\nFor example, imagine a robot learning to walk using this trial-and-error approach. It would first need to learn how to keep its balance, then understand how to move its limbs so it doesn\u2019t fall over, and then start learning how to move its limbs so that it can move from one point to another. As you can imagine, this approach requires a lot of training data, and obtaining this data from real, physical systems can be expensive and dangerous.\r\n
<\/h6>\r\nWhich is why simulation models are becoming increasingly crucial for developing autonomous systems and are especially integral to the success of a machine teaching approach. Simulation models built using Simulink and Simscape are perfect for modeling the wide range of environmental and operational conditions an autonomous system might encounter in the real world. The data generated by simulating these models in a variety of conditions is then used by the machine teaching algorithm, resulting in an autonomous system that is trained to perform in a number of scenarios.\r\n
<\/h6>\r\nOur partnership is focused on enabling the use of models built using Simulink and Simscape with Project Bonsai, and more importantly, scaling simulations of these models in parallel on Azure, which significantly improves training performance. This is done via the Bonsai Toolbox, which provides a Simulink block that can be added to your existing Simulink model to connect it to the Azure-based Project Bonsai platform.\r\n
<\/h6>\r\nThe Bonsai Toolbox can be accessed via the Add-On Explorer in MATLAB or File Exchange<\/a>. For scaling training, you can upload your Simulink model files to the Project Bonsai platform and parallelize simulations to improve training performance.\r\n
<\/h6>\r\nProject Bonsai is now generally available and you can learn more about the platform here<\/a> and Microsoft\u2019s solution for designing autonomous systems here<\/a>.\r\n
<\/h6>\r\nMathWorks and Microsoft have also collaborated on an example that demonstrates how a Simscape model can be used for training an autonomous system with Project Bonsai. We have trained a robot to balance a ball on a plate and navigate virtual obstacles on this plate using machine teaching.\r\n
\"\"<\/h6>\r\nWatch the YouTube Video here!<\/a><\/span>\r\n
<\/h6>\r\nTo download the Project Bonsai examples files, including the Simscape model, access the File Exchange submission here<\/a>.\r\n
<\/h6>\r\nTo learn more about building physics-based models of your system and concepts of reinforcement learning and machine teaching, click on the links below.\r\n
<\/h6>\r\n