By Frank Domoney
Professor Mousannif and I ran a six-day workshop covering the essential skills to build a highly automated factory. Digital Twinning of Future Smart Factories will require enormous amounts of computing capacity (see this article for more details) that successfully integrates Supercomputing and the Robotics stream of the European definition of Artificial Intelligence.
This coincided with the general availability of the NVIDIA Xavier embedded module and the launch of the EU AI initiative.
One of our objectives is to align Moroccan Engineering Education with the German Approach to Industrie 4.0 so there is a pool of readily employable, highly trained graduates in Marrakech to integrate with the European AI and Robotics market in Europe.

The biggest complaint I hear from students in Morocco is that the University Sector is so poorly funded that the students never get to build anything. This is borne out by fact that one of the biggest universities in Morocco hadn’t managed to buy a 3D printer yet.
EPCC at the University of Edinburgh published a set of instructions for the construction of a suitcase sized model supercomputer called Wee Archie. Included in these are instructions for the construction of a Five Raspberry Pi Cluster called an Archlet.
So we built three Archlets using the instructions that students modified themselves, and once we had three running at once we combined them all into a 36 core machine and ran the usual performance measurement tools on it.


We talked about the NVIDIA Saturn Machine and the Summit and Titan Machines at Oak Ridge National Laboratory. To discuss Supercomputing beyond Exascale we introduced Quantum Computing and we held a most interesting discussion of the possibilities opened up by this volume of computing capacity. One of the really rewarding moments was when one of the students showed how well read he already was on Quantum Computing.
On the second day we looked at how the Robots and Sensors will communicate reliably among themselves in the Smart Factory. We started with the excellent webinar by Lightbend, explaining the state transition model in Akka using LED arrays on the side of a five-Pi Archlet.
Students were already familiar with Spark and Cassandra so it was only necessary to introduce MESOS and Kafka to complete the SMACK architecture. While we had three interacting Spark notebooks available to demonstrate a Spark toolchain, we were unable to demonstrate the traffic source because the device gave a Java error. So we decided to leave this as a project for the future, and ran a Flink Tutorial instead. The participants were delighted to be able to monitor modifications to Wikipedia in real time.
We discussed the difference between Imperative Programming and Functional Programming and we tried out some of the transformations between the two systems.
After lunch we progressed to the Lightbend tutorials, which we hope to use these to set up a Lightbend Centre of Excellence in Marrakech to provide design and consultancy all over Africa.
The following day we covered the essentials of Deep Learning and the use of GPUs. We described the Architecture of the NVIDIA Pascal and the newest NVIDIA Volta along with the structure of the Xavier Embedded module, which had just been released.
We then spent a day having a great fun with Machine Vision on Jetson TX2 and talked about the Jetpack software and its various components. We only used the on-board camera on the device but managed to demonstrate not only static object recognition but also real time object detection.

Sadly we didn’t have a Xavier Development kit or compatible camera modules with us. Just after the workshop e-con announced e-CAM130_CUXVR – Multiple Camera Board for Jetson AGX Xavier, so maybe next time we will get to use it.
We watched the NVIDIA’s webinars on Xavier and Deep Streaming. It is our intention to set up a pair of software houses to exploit this technology, one of them for ladies only.

Sadly our 3D printer hadn’t arrived yet so we had been unable to print and assemble a robot. We walked through the basic robotics packages with ROS, Moveit, and Gazebo and the students discovered a cloud-based package integrating all of them. Many participants were already familiar with reinforcement learning so we watched the Siemens video from CeBit hon practical reinforcement learning with a cart pole.
As we hadn’t received out Xavier development kits we didn’t have access to NVIDIA’s Robot Simulation ISAAC. We did, however, discuss the syllabus of a Professional Robotics course based on Professor Lynch’s book “Modern Robotics: Mechanics, Planning, and Control”. We spent a happy afternoon building the BubbleRob and Hexapod robots with V-Rep. We will definitely build the real thing once we get the 3D Printers! Like the one in this video.


Future Steps
Once the NVIDIA Xavier development kits and camera units arrive we will build some vision-equipped robots.
To mitigate the issue of limited resources a local maker space similar to the Ipswich Makerspace will offer workshops allowing people to build their prototypes and 3D-printed robots.
We are building a centre of excellence in AI, HPC and Robotics in Marrakech so we plan to offer an opportunity for the recent University graduates and those who are self taught to upgrade their skills.
To this end we will run a Continuing Education Programme offering a number short courses to up-skill people in the Marrakech area.
We also hope to encourage more students to apply for the Chevening, DAAD and other scholarships. UK FCO offers Chevening Scholarships to promising candidates who have a few years experience under their belt and are recognised as leaders in their field to study a Masters degree so the Continuing Education Programme might offer a gateway to these.
DAAD offers scholarships to attend German Universities and the five hundred French scholarships offer opportunities in the French language for people who want to learn more.