Can we solve robotic piece picking challenges with 3D machine vision?
Robotic piece picking is part of automated order management used in warehouse logistics. A typical scenario involves picking several individual objects from containers (pallet, bin, tote, etc) to a final, target location (for instance a shippable box). A fully automated system uses industrial or collaborative robots (cobot) to fulfill these random, yet repetitive tasks.
The demand for robotic piece picking increases among international e-commerce companies that need to operate 24/7 continuously. With successful piece picking implementation, companies can expect productivity improvements and order accuracy at lower costs.
However, this simple chain of operations that seems easy for us humans, like detecting objects, picking, and placing them correctly, is still a challenge for robots.
Last year only, Amazon hired more than 400,000 workers to meet and handle online orders — the tasks that are not yet fully automated. It gives us an indication that there are still some challenges with automated piece picking solutions. This article covers three typical challenges with robotic piece picking and how to solve them with cutting-edge 3D vision technologies.
Challenge 1. Tiny, densely stacked, or randomly arranged objects
Detecting and picking unknown SKUs (products, parts, objects) of various shapes, sizes, and materials is one of the biggest problems in logistics. Unpickable objects often have the nickname “uglies” — it’s those objects that is hard to image and grip.
In piece picking, robots are often required to handle tiny, thin, porous, deformable, or highly irregular objects densely stacked with a little gap or piled on top of each other. The stacking and random placement cause 3D scanners and depth sensors with a wide baseline (the distance between the cameras or laser and camera) to miss objects due to occlusion! Smaller objects hide in shadows of other objects or the side of the container. Occlusion issues may result in lower accuracy and an increasing number of mispicks.
Zivid 3D cameras are designed for covering typical piece picking objects in industrial automation environments. The cameras have a small baseline making it easier to distinguish features smaller than 5mm, with minimal occlusion. Its powerful 2.3 MPixels image sensor ensures reliable object recognition and separation of boundaries even for very tiny and thin objects.
Challenge 2. Shiny, reflective, and plastic-wrapped objects.
Handling shiny, reflective, and plastic-wrapped objects are common challenges in machine vision applications. They can cause point cloud artifacts and negatively affect the robot’s ability to detect, pick, and place items correctly. For example, artifacts like inter-reflections can lead to a heavily distorted point cloud with floating groups of points, “ghost planes,” and a seemingly random scatter of noisy points.
Over the years, Zivid has developed unique machine vision technologies like 3D HDR and Artifact Reduction Technology (ART) to ensure excellent suppression of imaging artifacts from reflections, interreflections, specular highlights, and high transitions.
ART includes a reflection removal filter called ContrastDistortion, which improves the 3D image quality from primary underlying error sources. With Zivid 3D cameras, users can also benefit from the Stripe engine, which adds optimal patterns for capturing point clouds of shiny or reflective objects to see an immediate improvement in image quality.
Challenge 3. A wide variety of objects.
Capturing a broad range of objects is essential in piece picking scenarios where there is usually a mix of objects in a bin. It requires a machine vision system to recognize objects with similar shapes, colors, and material fast and separate them for proper and place operations. There are many 2D or low-cost 3D vision cameras that are not necessarily optimized for piece picking applications. Robots might detect objects with such a solution, but they cannot pick them every time due to scaling, rotation, or translation errors.
Zivid 3D cameras ensure that robots can detect, pick, and place objects faster than typical stereo 3D cameras used in piece picking today. With Zivid 3D, robots can capture typical piece picking objects <100 ms for low to medium dynamic range and < 300 ms for high dynamic range scenes, including shiny and dark absorptive. The unique combination of native color and high dynamic range enables a wide object coverage, including ceramic, metal, cardboard, and wood with a low occlusion. The Zivid cameras are also designed to work with all kinds of grippers (suction, vacuum, fingers stiff/compliant, soft) and meet strict dropping requirements.
Despite different use cases, the goal is the same for piece picking. Customers want automation that allows flexible picking solutions with high-accuracy, reliability, and speed. It is crucial to choose the right machine vision system to meet the business requirements by testing all the piece picking scenarios addressed above.
Zivid is a pure-play company specialized in 3D machine vision. Its focus is to provide hardware and software solutions for easy 3D application development in the world of automation. You can visit this page to learn more about solving piece picking challenges with 3D machine vision.
