Enabling robots to pick unknown objects from bulk

Robotic technologies have been playing an important role in the supply chain industry for decades, increasing efficiency across a variety of operations. More recent advancements are transforming the traditional role of robots from assisting humans in repetitive tasks to replacing them in more dangerous and complex ones, while improving overall performance. A great example is computer vision, a technology applied for decades, yet only recently realising its full potential. With artificial intelligence making a more prominent appearance in the world of robotic automation, the application of neural networks for computer vision has also increased drastically, giving life to new solutions for challenging business cases. 

This transformative progress was recently sped up by the global pandemic, which radically impacted the supply chain and logistics industry, making the need for robotic automation more urgent than ever. With almost 80% of labour in warehousing being dedicated to picking and packing, companies are gradually investing more resources in logistics automation. However, many are still struggling in applications involving items of irregular shape, unaware that robots have become more capable of coping with variation. Indeed, there are robust solutions in the market for applications like item picking, parcel handling, depalletising, and truck unloading, that enable robots to handle an unlimited number of SKUs with great efficiency, while facilitating cost and speed optimisation. The secret lies in the use of neural networks in computer vision, where deep learning algorithms enable the automated handling of unknown objects varying in shape, size, colour, material, and stacking, being picked from (conveyed) bulk. 

Instead of acquiring a proprietary picking cell, that most piece-picking companies in the market offer, integrators now have the freedom of choosing the best hardware for their picking cell, in the form of plug-and-play modules. A good example is Fizyr’s modular software product that integrates smoothly with any camera, robot and end-effector, allowing companies to choose the best equipment to meet their needs at any time. The algorithms provide over 100 grasp poses each second, including classification to handle objects differently, but also perform quality controls to detect defects to prevent damaged items from being placed on a sorter. 

This technology is already solving big challenges for companies operating worldwide in online retail, warehousing and parcel services. Using such a computer vision software, companies are provided with all relevant information about segmentation (including white on white flats or reflective bags), classifiers of type of parcel (including box, bag, envelope/flat, tube, cylinder, deformable, etc.), determining outliers or non-conveyables (i.e. damaged goods), best possible grasp poses in 6 DoF, and multiple ordered poses per object. These algorithms also allow sensors to deal with closely stacked or overlapping objects, highly reflective items and apparel in polybags, black-on-black flats, as well as transparent objects.

The key to tackling both the uncertainty created by demand fluctuations and the large variation in items and parcels is to move towards operation automation. This will allow companies in the logistics industry to improve conditions for their employees working in repetitive and dangerous positions, achieve cost reduction and, given the circumstances created by Covid-19, minimise human contact as much as possible. System integrators worldwide have already started acquiring and deploying high-quality solutions like Fizyr’s, to build cutting-edge automated picking cells and have robots assist them 24/7 in dangerous or repetitive tasks.

Herbert ten Have, CEO, Fizyr

For more information, visit www.fizyr.com