November 28, 2017      

Ocado Technology, the research division of online grocer Ocado Group Ltd., is no stranger to robotic grippers and grasping techniques. Ocado has been working with European partners to develop soft manipulators, including SoMa, to help fulfill orders in its e-commerce fulfillment centers.

One of Ocado’s main goals is to develop a robot that can reliably pick the 50,000 items it offers and help fulfill an average of 260,000 orders per week. While it is making “significant progress” on SoMa, Ocado today unveiled a suction-based picking system that offers a “more immediate in-house” solution and uses off-the-shelf components.

The new picking solution, which hasn’t been named, is “conceptually simple in its design,” consisting of a suction cup on the end of a Staubli RX160 robot arm. There’s a pipe that connects the arm to an air compressor, allowing the robot to pick up Ocado’s SKUs regardless of their shape and deformability.

The robot uses a 3D vision system to locate an area on the object that’s big enough for the suction cup to latch onto and create an airtight seal. Once that designated grasping point is located on the object, the robot arm lowers the suction cup onto the item. Watch the video below of the system in action:


Graham Deacon, Ocado’s robotics research team leader, wouldn’t specify the picking rate for the robot. He simply said it is “comparable to a person.” He said repeatability is the main reason Ocado wants to automate bin picking inside its fulfillment centers. “We know it’s going to do what it’s supposed to be doing.”

Along the way, the system verifies that the picked item is the correct product, and determines the optimal orientation to rotate it to, before placing it into the crate. The robot can pick up items that weigh up to 300 grams (0.66 pounds), which doesn’t sound like much when compared to other suction-based picking robots. But Deacon said the system isn’t limited to the suction cup it currently uses. “It was convenient,” he said. Using a different suction cup would likely increase the maximum weight the robot could lift.

IAM Robotics’ Swift, for example, is a mobile picking robot that uses suction and can lift up to 15 pounds. IAM also creates models of the SKUs that Swift picks, saying its Flash product dimensioner “in less than one second” records the barcode, 3D dimensions, weight and a high-resolution image of any product” it scans.

Because Ocado’s new system uses suction, porous or corrugated items, Graham said, would be challenging to pick. Graham said built-in sensors prevent items from being crushed or damaged during both picking and placing. The system also ensures items are only released if they can be placed without protruding from the delivery crate.

Ocado Bin Picking Robot

Ocado is keen on developing technology in-house to own the IP, but it isn’t keen on modelling every SKU in its catalog, as many of today’s picking systems do. But Ocado said “we knew there must be a quicker and more efficient method.” So it took a “model-free” approach when designing this system.

“The short-coming of a model-based approach is that you can only pick up objects that you have a model,” said Deacon. “It’s quite challenging to build 50,000 different models. Our model-free approach just looks for things in its environment – for a patch on each object that can accommodate the suction cup. This is more versatile and less restrictive than a model-based approach.”

The image shows the initial concept design of the robot pick station. The robot picks up items from storage crates (green) at the pick station and places them in delivery crates (red) for customers.

Starting in January 2018, Ocado will start to roll out this new picking system in its customer fulfillment centers to expand the number of SKUs it can handle.

Deacon said Ocado is not against selling this system to non-competitors. “If it was worth our while, we would sell it to outsiders. The customer needs to be in a completely different section, such as automotive, but why not?”

Ocado is also working on SecondHands, a Horizon 2020 Project, to develop a robot assistant that is trained to understand maintenance tasks so that it can help maintenance technicians performing routine and preventative maintenance.