Automating the operations in warehouses, on production lines and various other industrial areas aims to improve productivity, remove bottlenecks in the process and deliver a higher quality outcome. Industrial automaton uses control systems to carry out this process with the employment of robotics, computer systems, and artificially intelligent systems to increase the performance and outcome of the process.
Until now, this is work that has been carried out by human employees, so why switch now? With the advent of the IIoT and the fourth industrial revolution, operation costs are lowered, safety is improved, and employees are freed up to work on tasks that require more care and skill that these replacement robots are unable to do. The margin for error is also hugely reduced, as automation alleviates any human errors encountered through tiredness which results in mistakes and poor outcomes. Whilst this is the overall goal of industrial automation, it's still not perfect. It's still early days, so it's vital that procedures are monitored. Working to provide visual inspection for manufacturing and logistics with the goal of removing errors and bottlenecks is Aquifi. Aquifi combines 3D computer vision and AI, providing services such as object dimensioning, identification and inspection for defects and anomalies. Carlo Dal Mutto, CTO, coordinates the development of the technological foundations of the company, spanning from hardware design and depth sensing to computer vision and high-level inference based on AI.
At the AI in Industrial Automation Summit in San Francisco this June 28 - 29, Carlo will be sharing his most recent work in the space, and in advance of his presentation, we spoke to Carlo about Aqufi and what we should expect to learn in his presentation.
The heritage of Aquifi lies in the strong computer vision capabilities of our team, with a particular focus on 3D reconstruction. Application of AI to 3D data has major benefits, in particular allowing for better classification and detection performances, with respect to standard image-based AI while requiring less training samples. This constitutes a major benefit towards the application of such techniques in different applications in manufacturing and logistics. At the inception of our AI endeavour, we were considering different applications from consumer-facing, e-commerce and b-2-b focused. After few interactions with potential customers, it was clear the current the logistics and manufacturing fields were undergoing a deep revolution, driven by the Industry 4.0 paradigm and AI, in which however 3D computer vision could have played a fundamental role.
There are two main components in the system: (1) an hardware setup constituted by one or more of our sensors, which include a depth camera, a color camera and other sensors, and (2) a software pipeline that processes the acquired data in order to provide a 3D model of the acquired objects, and use such models for identification.
The identification pipeline is itself characterized by 3 steps:
a training phase, in which a general AI model is trained in order to provide the foundations for identification. Such training phase is continuously happening and is application-independent.
a one-shot learning phase, in which the system acquires one or more 3D models for each instance to be identified, using the trained general AI model to compute a unique signature for every instance.
a deployment phase, in which the 3D model of each object to be identified is acquired, its signature is computed by the training general AI model and it is compared agains the signatures of the instances computed during the one-shot learning phase in order to perform such identification.
Regarding our application, there are two major benefits that are pursued: the reduction of errors in manufacturing and logistics and the reduction of workers fatigue.
Starting from manufacturing and logistics errors, their reduction allows on one hand increase customer satisfaction and on the other hand to preserve the environment from unnecessary pollution and contamination due to the lack of early-catch of such errors. In fact, let us consider the case in which a wrong product is delivered by e-commerce. Such product is likely going to be returned, and then the correct product is going to be shipped. This leads to three product shipments, with unnecessary pollution caused by such wrong shipment. Even worse is the case in which the object is defective because it also has to be disposed of.
On the workers fatigue, we position our technology as an inspection tool that carries on repetitive visual inspection tasks, which have shown to be fatiguing if not dangerous for workers. Our goal is to perform information aggregation such that workers are involved in the loop only once errors or potential errors occur.
The major problems we are trying to solve with AI are identification and inspection for defects and anomalies. The main challenges we are currently facing are related to the variability of the different situations in manufacturing and logistics and in the design of a unique system that is able to cover a relevant set of them.
Similar technologies can be applied in many other contexts, such as retail analytics and autonomous robots. In particular, the field of autonomous robots, application to sortation and picking are very interesting and are definitely in our roadmap.
Yes, I would advise a career in AI. I believe AI is a fundamental technology that is transforming and will continue to transform our society in the upcoming years, and several interesting discoveries are going to happen around it. In terms of key skills, I believe that strong math and statistical foundations are needed, as well as the capability to clearly specify and scope the problems that are going to be solved.
Moreover, I am really convinced that anyone that deals with AI have to develop an ethical consciousness to prevent unintended consequences of the application of such powerful techniques.