![\"unsupervised-blog-visuel1\"](\"https:\/\/sprint-project.com\/wp-content\/uploads\/2019\/05\/unsupervised-blog-visuel1-300x138.png\")
<\/p>\nOver the last two years, my co-founder Alexis Theallier and I, Cl\u00e9ment Jambou, have been working on the design of robots dedicated to warehouses in order to reduce the arduousness of the boring tasks linked to moving products. Behind this concept lies a desire to bring robotics to the heart of the supply chain industry to replace humans. We are convinced that robotics has a role to play in the major challenges of the 21st century, in particular to free humans from these tasks and make these logistical tasks more efficient, ecological and economical. The ability of robots to perform tasks autonomously is our company's ultimate vision: unsupervised.ai<\/p>\n
When we talk about autonomous robots, it is clear that there is a great disparity between the marketing messages and reality. The observation is obvious, if we stick to these announcements, autonomous robots already exist and it is an accomplished technology. But when we talk about autonomy, it is crucial to qualify it precisely. This principle has been applied in autonomous cars with different levels of autonomy ranging from level \u201c1\u201d: driving assistance, to level 5, where the vehicle can operate in all conditions (night, snow, etc.) on any kind of road.<\/p>\n
\u00a0<\/p>\n
<\/p>\n
The reality in the warehouse is very different. Most of the current robots deployed are line-following robots which stop when an obstacle is present; they are \u201cautonomous\u201d but under very specific conditions. These robots equipped with magnetic or visual sensors have little on-board intelligence but can be very effective for certain specific applications. In particular, they are increasingly deployed on assembly lines in the automotive industry.<\/p>\n
A higher level of autonomy consists of deploying robots in constrained environments, such as an area delimited by a fence or ground markings. This time these vehicles are able to move (sometimes very quickly) and find their way in such an environment which has been built around them. This type of deployment recently led to the success of Kiva (the largest acquisition in robotics was $775M by Amazon), or even impressive distribution center demonstrations from Alibaba. These robots, although deployed massively (>100,000 robots by Amazon) are not yet very present in Europe. Concerning their autonomy, these robots require building the warehouse around the solution and do not work with human operators at their side. The parallel for the autonomous car would therefore be the construction of a road dedicated to their operation. This does not detract from the technical prowess of these vehicles and the optimization algorithms for their fleet management, but it is unthinkable for a factory or warehouse combining operators and machines to imagine such solutions for all their logistical tasks.<\/p>\n
![\"unsupervised-blog-visuel2\"](\"https:\/\/sprint-project.com\/wp-content\/uploads\/2019\/05\/unsupervised-blog-visuel2-300x172.gif\")
<\/p>\n
It is questionable whether the interaction of robots with humans is really necessary. After all we could imagine warehouses or factories, where only machines interact with each other and where the difficulties linked to security and the interface between robots and operators do not arise. But the reality is far from what science fiction promises us in this sense. A large number of tasks, even simple for a human, are today unthinkable to automate even in the most advanced artificial intelligence laboratories. Certain tasks that are very simple for us, or even a child, are extremely difficult for a machine. In particular, the field of handling and gripping in factories and warehouses is unattainable (apart from in certain very specific cases equivalent to level 1) technologically. Amazon has notably developed a robotics competition<\/a> to help take automatic packets which have failed to find a winner for several years. Their robots move the shelves, but it is an operator who must recover them<\/p>\n \u00a0<\/p>\n The market therefore expects robots capable of operating in complex environments, alongside human operators. At Unsupervised we are therefore working on collaborative robots, which can map a new environment and move around in it, even if it was not designed around robots. Our robots can detect operators and avoid them. They adapt their speed according to the environment and can in particular be programmed for cohabitation zones and zones reserved for robots.<\/p>\n \u00a0<\/p>\n The first step in deployment is to map the location. We want to remove any additions to the environment specific to robots and conversely allow robots the opportunity to learn from their environment. This technology is a category of algorithm called SLAM (Simultaneous Localization and Mapping) and consists of the creation of a map of the environment (see animation) based on the robots' sensors. Indoors, the robots do not have GPS. Therefore, to find their way, they use their sensors (notably Laser and Camera) to identify points of interest and make a map. Once this map is made, at any time the robot tries to compare what it \u201csees\u201d to its internal map to deduce its position. In this sense it is very similar to the way in which a human finds his way in an environment that he discovers (by noting exits, and landmarks).<\/p>\n \u00a0<\/p>\n Once this mapping has been carried out, it is visualized on our application and annotated. This allows users to define points, routes and areas of interest for robots. This data can then be used to give simple missions to the robots (for example a round trip between 2 points) or more complex missions by interfacing with other software. For example, for a package sorting application for the postal service, a bar code is scanned which allows you to give and separate packages into several groups by destination. This action of scanning the package can therefore order the robots to do this task. Our robots can even interface with WMS (Warehouse Management System) type software which constantly tracks the status of a warehouse (SAP, Manhattan, etc.) and therefore automate tasks thanks to this monitoring.<\/p>\n \u00a0<\/p>\n \u00a0<\/p>\n \u00a0<\/p>\n The main advantage of these systems is their flexibility. It is very simple to add a new robot to a fleet since the robots share their data and therefore share their mapping information. The fleet manager allows you to optimize the movements of robots and the tasks assigned to them. In particular, in order to operate constantly, the robots have autonomous charging bases which allow them to recharge on their own when there are fewer tasks to perform. This type of system is therefore particularly suited to markets with seasonality (such as e-commerce or production). Finally, this also allows us to reduce the risk for our customers, thanks in particular to a rental model. This rental model only makes sense because installation costs are negligible due to the adaptability of our robots.<\/p>\n The current technology designed at Unsupervised allows our robots to be better than our competition on the market thanks to on-board artificial intelligence and the techniques that come from the autonomous car that we board. On the other hand, there is still a long way to go for this type of robot to behave correctly in any environment (in the city for example) and potentially without requiring a mapping or remote control step. The problem of autonomous navigation is complex and current technology does not yet unlock the equivalent of \u201clevel 5\u201d of autonomous cars. We must therefore be careful when analyzing these technologies. At Unsupervised, we see this as an opportunity to build a leading player in artificial intelligence and robotics in Europe.<\/p>\n![\"unsupervised-blog-visuel3\"](\"https:\/\/sprint-project.com\/wp-content\/uploads\/2019\/05\/unsupervised-blog-visuel3-300x226.png\")
<\/p>\n\u00a0<\/h5>\n
Artificial intelligence to build collaborative robots<\/h2>\n
![\"unsupervised-blog-visuel4\"](\"https:\/\/sprint-project.com\/wp-content\/uploads\/2019\/05\/unsupervised-blog-visuel4-300x169.gif\")
<\/p>\n![\"unsupervised-blog-visuel5\"](\"https:\/\/sprint-project.com\/wp-content\/uploads\/2019\/05\/unsupervised-blog-visuel5-resize-300x169.gif\")
<\/p>\n![\"unsupervised-blog-visuel6\"](\"https:\/\/sprint-project.com\/wp-content\/uploads\/2019\/05\/unsupervised-blog-visuel6-300x200.gif\")
<\/p>\nMore flexibility<\/h2>\n
![\"unsupervised-blog-visuel7\"](\"https:\/\/sprint-project.com\/wp-content\/uploads\/2019\/05\/unsupervised-blog-visuel7-300x169.gif\")
<\/p>\n