Artificial Intelligence: Reference Story Gestalt Robotics
The Deep Tech Field of Expertise in Artificial Intelligence: Reference Story Gestalt Robotics
KEEPING AN OVERVIEW – WITH GESTALT ROBOTICS EPIC SYSTEM
Gestalt Robotics efficient pipeline for image classification – or EPIC system for short – can instantly detect and locate a single object in a warehouse of thousands of objects, and autonomously learn new objects given just a handful of examples. Deep learning and data efficient AI technologies make this possible.
Life is usually pretty hectic in company stores and warehouses: raw materials and components are constantly being delivered while finished products are brought in for temporary storage to await collection by shipping firms or customers. Yet warehouse staff still need to keep a watchful eye on the current stock status and flow of material because it’s the only way to avoid mistakes and ensure the seamless flow of the supply chain. Add to this an ever-changing commodity and product portfolio which means that new objects must be permanently accounted for. Such requirements pose a particularly big challenge to small and medium-sized enterprises (SMEs) with a low level of digitalisation.
Deep Learning and Data-efficient AI in Action
With Gestalt Robotics EPIC system SMEs now have an effective and efficient way of monitoring their warehouse stock. Integrated in 2D and 3D cameras, their AI solution can instantly recognise colour images of all kinds of objects such as pallets, boxes, products and components as well as instantly identify objects it doesn’t know. As a rule special deep-learning technologies were used for this, yet such technologies are data hungry which for a long time made them too inefficient and expensive for industrial use – until Gestalt Robotics presented the first prototype of the EPIC system at the GPU Technology Conference in October 2018.
The EPIC System in Warehouses
The EPIC system is so innovative because it can learn from a tiny number of images without any need for time-consuming training. Generally, about five shots of an object taken from different angles are more than enough. EPIC uses few-shot-learning to transfer images to a feature vector. As this element is highly memory-efficient, it cannot be influenced either by ambient light effects or the spatial position of an object or by any partial covering. This is because single shots of pallets, boxes etc. in different settings have similar feature vectors. When deployed in warehouses and stores, the EPIC system integrated in the cameras identifies objects by computing their feature vectors and uses comparison algorithms to compare then with the feature vectors already in the database.
Gestalt Robotics ships the EPIC system both as a customised solution for industrial users and as a software developer kit and web service for sensor manufacturers and integrators. Both versions meet the highest quality standards as is shown by the bestowal of the Deep Tech Award to Gestalt Robotics for its EPIC system in April of this year by a prestigious jury of experts.
MORE REFERENCE STORIES FROM THE IT FIELD OF EXPERTISE IN ARTIFICIAL INTELLIGENCE
Birds on Mars supports Companies with AI Thinking
Birds on Mars supports companies in the development of strategies, teams and applications across the connecting line between human and artificial intelligence; creates a new form of consultancy along the way, and with its Artificial Muse investigates the future of creativity and work.