This week we’ll be exploring the intricacies of each picking robotics in warehousing as well as the current state of the art in this technology, specifically the gripper technology currently in use. Before I decided to get into warehousing automation I was involved in more traditional automated systems in the process automation and manufacturing industries. Factories in these industries have been using robotic manipulation and grasping for decades and this technology is quite mature. But these traditional robotics systems are relatively closed loop systems just like the traditional ASRS systems we’ve discussed in recent articles.
Source: fanuc.com
Industrial robotic arms are able to pick and place pre-configured items onto pre-configured locations ) at very high speeds but with limited variability. These systems are usually programmed using PLC control meaning that every variable needs to be hard-programmed into the control system. The rise of the software PLC or ‘soft’ PLC has enabled faster programming and the change to a configuration environment (vs the traditional coding environment). This helped to empower robotic programmers to build control systems that have a smooth interface with a user-friendly software platform for increased usability and flexibility. However, the cost and complexity of industrial robotic systems has meant that warehouse/fulfilment use-cases have remained in the domain of R&D departments. Add to this the fact that these robotic arms need to operate behind safety fencing in Robot-only areas, you can see why designing them to accommodate the high-variability requirements of each picking in a warehousing has been such a challenge. Although these traditional systems laid the foundations for each picking, it was necessary for a new wave of technology to rise to really help us get there.
Collaborative Robotics; the secret sauce
Where traditional closed-loop industrial robotics were limited to robot-only areas, the new breed of collaborative robotics are able to work alongside a human workforce. This is done through an inertia-reducing design with built in safety functionality in the form of a sensitive shell which stops or reduces robotic movements if any impact is felt. An increasingly accessible feature in a lot of these ‘cobotic’ systems is the ability to use a teach mode where the robotic arm is able to be physically manipulated in order to ‘learn’ how to perform a movement it has never performed before. At ProMat 2019 I had the opportunity of testing out a couple of different teach modes on UR, Fanuc and Hitachi robots and the ease of use with these systems is truly remarkable. The proliferation of these robotic systems has served to supercharge development of the vision and grasping technologies required to increase the versatility and autonomy of these solutions.
Source: australiseng.com.au
AI powered Vision
Vision systems are an important part of traditional industrial robots. In these applications vision systems are designed to perform in a structured environment where the orientation and shape of an object presented to the robot is pre-defined and the role of the vision system is to guide the robot gripper to and from the pick and place robot often without safety sensors which are an integral part of cobotic systems. Development into Vision technologies for robots has been advancing rapidly over the last decades as accuracy and speed along with quality auditing became key selling points of industrial robotic systems. I remember being wowed by a solution developed by Scott Automation and Robotics where an industrial Robot was used in meat processing to maximise the yield of meat off a carcass. Basically, an X-Ray vision system would scan the carcass and determine the best way to cut it for a maximum yield before the Robot would perform a ‘samurai-like’ maneuvre with a large blade and slice away. You can see the video here. https://www.youtube.com/watch?v=KhZ2jl-exBs This is a great example of how far industrial robotic systems have come.
The maturation of machine learning and deep learning algorithms in the wider programming community has helped to develop robotic systems which are able to handle a large variety of (often unconfigured) shapes and sizes. Put simply, these artificial intelligence algorithms are able to ‘learn’ from previous grasping and manipulation scenarios and make an ‘educated’ guess on how to grasp a previously unhandled object, just like Humans! Using different types of sensors and cameras, these systems work on detecting key parameters in an object and comparing them to a database to come up with the best orientation and surface from which to grasp. In some cases, it is even possible for a system to learn by watching human operators perform their handling tasks or having them ‘teach’ the robot for a base range of products as is possible with cobotic arms. There are now even simulation software platforms which enable training of robots through the simulation of millions of grasps in a virtual environment. But there was still one final piece of the puzzle missing where humans outperform robotics and that is in adaptable grasping and manipulation technology. As Humans, we are masters of dexterity, but the robot is no longer too far behind.
Source: abb.com
Grasping and Manipulation
Any one of us is able to walk up to a box of objects, pick out any object and place it on any surface within our reach without needing to think about it too much. But getting a robot to do this is actually quite difficult. Apart from the programming/configuration effort and vision technology required to enable a robot to ‘think’, ‘see’ and ‘learn’ the act of grasping an object is also challenging across a variety of products. Promat 2019 was held alongside Automate 2019 and it was clear that the similarities between the traditional industrial robotic systems and the rising materials handling systems created many synergies. The Automate portion of the exhibition was largely about industrial and collaborative robotics being showcased and the Promat portion was putting these solutions to work in picking/fulfilment environments. It truly felt like the collaborative robot was centre stage and coming of age. I identified 3 key types of robotic grippers being used to grasp and manipulate: Vacuum gripper, parrallel jaw gripper and a hybrid vacuum parallel gripper.
Source: sclaa.com.au
Vacuum Grasping
This is the simplest method of picking an object as it uses a vacuum controlled suction cup to attach to a flat or ‘control’ surface on an object to lift it and enable placement into the outbound vessel. Vacuum gripping has been an integral part of industrial robotic handling systems since the beginning and since this technology is quite mature, it was the obvious starting point for each picking robotics. Almost all of the robotic manufacturers and materials handling system providers have an each picking solution which uses this proven technology to pick and place from the product tote into the order carton for fulfilment. This example from Dematic has been chosen by Drakes Supermarkets as one of the first robotic Each-Picking solutions to come to Australia for order fulfilment and I’m excited about the opportunity to commission one of these solutions in the near future. But this simple design is still quite limited to a select range of products within the larger Stock Keeping Unit (SKU) profile of any organisation.
An innovative solution I witnessed at Promat using this technology was from Kindred AI who were using this proven technology in a great way to outsort batched orders for fashion/apparel ecommerce fulfilment. Check out this system here. The XYZ robotics and IAM Robotics each picking and placing solutions also use this vacuum grasping method proving that this simple solution is still a great way to pick and place. We’ll be looking further at these and other piece picking solutions in the coming weeks.
Source: https://engineering.eckovation.com/pick-place-robotic-arm/
Parallel Jaw Grasping
The next iteration of piece-picking robotic iteration is to mimic the grasping action of human fingers. Again, something we humans can do intuitively and easily, is quite hard to replicate with robotic gripper technology. There’s 2 key aspects to this challenge. The first is determining the amount of clamping force required by parallel jaw grippers to successfully pick up an object without crushing or damaging it. This force-feedback loop is relatively complex as the changes in gripping force can be quite minute. There has been quite a lot of development in this area over the years and advances in torque sensors in servomotors have helped to make this grasping method a possibility.
Universal Robotics is one of the leading suppliers of Collaborative robots and their platform has been used by many gripper manufacturers to develop innovative parallel grippers for picking and placing. Their website provides a great overview of the grippers available for their collaborative robot arms from specialised gripper (also known as ‘End Effectors’) suppliers. OnRobot was one of the End Effector suppliers who caught my eye at Promat with their interesting 2-finger gripper solutions for pick and place. You can see the relatively simple designs they’ve developed with their RG2 and RG6 Grippers. Robotiq, another key player in this market has also released a 3-finger adaptive robot to increase the versatility and flexibility if these grippers across a larger pick/place product profile. You can see their gripper solutions here.
Source: Right Hand Robotics
Hybrid Vacuum-Jaw Grasping
“Porque no los dos?” (But why not both?[Spanish]).
Where most gripper solutions seem to focus on either a vacuum-suction cup or finger-grasping methodology for picking and placing, there is a new wave of solutions coming to the market with a hybrid of both of these technologies. The leading player in this market is RightHand Robotics with their innovative 3 finger gripper with suction cup design. This solution was the most popular each-picking gripper I saw at Promat 2019 and was being used by at least 5 different materials handling suppliers for completely unstructured picking and placing for fulfilment applications. One of the impressive features of this solution was the ability to ‘flick’ objects in order to change their orientation and present a better surface for grasping from.
The Bastian 3 fingered gripper (mentioned in previous articles) was also quite interesting as it used a 3-finger design with a suction cup on each of the fingers. I’m not convinced that more suction cups is necessarily better, but an interesting approach nonetheless.
There are other variation of these grippers worth noting such as the ‘soft-finger’ grippers made famous by Soft Robotics which are now into their 2nd and 3rd generation releases enabling soft grasping of fragile objects such as fruit. Their end-effector utilises a vacuum driven finger grasping design for increased dexterity with a ‘soft-touch’ which comes from the plastic/rubber design of the gripper itself. Soft also released a 3-finger design at Promat which was very interesting. Further to this, there are also now dual arm robots which mimick human behaviour in that they can pass an object from one gripper to the other or provide assistance with the second arm. We’ll take a look at some of these solutions in the coming weeks. We’ll also look further into the manipulation and placing side of the technology which has been developing at a rapid pace.
Jeffrey Triantafilo is a Warehouse Improvement Engineer in Intralogistics for Fuzzy LogX an Australian boutique consulting firm, based in Inner-Western Sydney, focused on helping Warehousing and Distribution Operations improve and stay competitive in today’s ever-changing supply chain environment. If you’d like to have a chat with jT or one of his colleagues about improvements in your operation, click here.