A swarm of robots moving like rooks on a chessboard, was how Paul Clarke, Chief Technology Officer (CTO) of Ocado Technology, described the Andover-based factory’s picking and sorting system.

The box-shaped giants are the latest addition to the company known for its online grocery services. These robots glide across the grid or ‘hive’ which is made up of stacks of boxes which are lowered and raised in narrow shoots according to which is needed. The robots act as a goods-to-person system, bringing crates of products from storage to pick stations where personal shoppers – either human or robotic – assemble customer orders by transferring products from storage into customer bags. The robots then take the crates back into the grid or to another pick station. They are also tasked with replenishing the storage structure by grabbing crates full of products from an area on the side of the grid and placing then back into the grid.

Each bot is exactly the same, meaning that one can be easily replaced by another if, for example, maintenance or updates need to be made. The robots can also collaborate to carry out tasks which require more than one bot, for example moving particularly heavy loads.

Currently, Andover has around 440 robots in its ‘swarm’, which are capable of accelerating at 2m/s2 and reaching speeds of 4m/s, but the aim is to increase capacity to 1,100 with the hope that at full capacity the facility will be able to process 65,000 Ocado orders ever week.

The swarm was born from the expansion to overseas operations. Initially, Ocado Engineering was a systems integrator, it would purchase third party hardware from logistics companies and install them in the warehouses. While Ocado Technology would rewrite the whole software stack to optimise for grocery operations.

Clarke said it proved successful in the first two UK warehouses (Hatfield and Dordon) but expanding its operations meant this solution was not scalable. For OSP, it needed to control the supply chain and to design a solution that could easily be installed and scaled in any geography. In response to this, it expanded the engineering team and added a product development team to design a new warehouse technology.

“We want to be able to sell this system into territories where online grocery is more nascent,” Clarke explained. “Rather than building a huge facility in one go, a company may wish to start with a smaller one which can be expanded. The grid can be partially populated and when the need to grow arises, more robots and crates can be added.”

In terms of the software stack, the low-level software that controls the robot is written in C, while the high level overall control software is written in Java. The system uses OpenStack to provision an internal cloud infrastructure where Ocado uses Kubernetes to create containers for the various applications it needs to run. Log data is stored in Ceph and everything is streamed into Google Cloud where Ocado uses BigQuery for analytics.

To talk to the robots, Ocado Technology designed what it claims is the “world's densest mobile communications network based on 4G technology”. According to Ocado, the communications protocol guarantees a connection 10 times a second to each of the 1,000 robots per base station – all working within a 150-metre radius.

The communications network is said to be the world's first deployment of unlicensed 4G spectrum for warehouse automation.

The wireless protocol uses orthogonal frequency-division multiple access (OFDMA) which is used by LTE for the downlink air-interface. However, Ocado has replaced the media access control (MAC) layer with its own implementation and entirely removed the upper layers of the 4G cellular offering, which were not relevant for this particular project.

The system has the notion of a resource map where the frequency and time allocation for each node is described. Ocado’s wireless protocol's map is fixed for each client as it adds it to the system vs LTE or WiMAX where some element of scheduling is involved.

In the future, Ocado is looking to add to its robotic-based facility and are currently working on a range of picking solutions, including SoMa which aims to perfect human-like grasping and Second Hands, an intelligent robot which will learn from humans and devise ways to assist them.