Can the logistics sector reinvent itself with artificial intelligence?
April 15, 2019
How well do you really know the world of logistics? Since Big Data has been on the scene, everything has changed! Gone are the days when around 2.9 million trucks, a thousand freight cars or trade ships simply traveled from A to B. Gone are the days when the supply chain was “only” a route from the manufacturer via distribution and trade to the end customer. That's because we have Big Data now! Thanks to optimization methods such as Deep Learning, the industry is not only able to learn more about it but – to put it simply – to reinvent itself.
Deep Learning has already been successful in medicine and in the automotive industry. One of the reasons for this includes Big Data, which is a term which we constantly encounter in the context of artificial intelligence (AI), machine learning or neural networks. Anyone wishing to understand Deep Learning must first gain a knowledge of artificial intelligence. AI can display weak or strong characteristics. Very strong AI strives to teach machines to do everything that humans are capable of. With a view to our sensory perceptions such as hearing and sight, this is entirely realistic. When it comes to taste or feel, this becomes more difficult.
Weak artificial intelligence: always present
So-called weak AI focuses only on individual capabilities which are transferred from humans to machines. Smartphones and laptops have weak artificial intelligence – skills such as text, image or language recognition are established. Even recognition of stop signs and traffic lights are an element of weak artificial intelligence within the automotive sector. The basis of everything (and often used as a synonym for Deep Learning) are artificial neural networks which are modeled on the human brain with neurons and synapses. Synapses are connections – within the neural network, they are almost like the input cable which transports the signals to the cell body and transmit the outgoing signal they produce to the next nerve cell. If cells which are bound together are active in conjunction with one another, the synapses will strengthen. Unused synapses will weaken. In this way, learning once again activates a number of cells which are linked together, and the connection between them gradually becomes stronger. Result: the neural network of a thinking, learning human.
Neural networks: never tired of learning
Deep Learning is the ability of a machine to recognize such structures independently, to evaluate this recognition and to independently and permanently optimize this using several progressive and regressive throughputs. The precision of recognition and the benefits of results become even greater until they present almost the same intellectual capabilities as are present in humans. Deep Learning may, for example, recognize images which are not subject to any rules. Yes, even those for which rules exist but which are not well known. The neural network is also fed with comparable information from Big Data. For example, this could be images with packages. Images of large, small, brown, green packages and so forth, the images of which are scanned into the network. During the learning phase, the network will repeatedly provide a notification of whether it has recognized a package as such, or not. Depending on the response, the network changes the connections between the neurons – it “learns”.
The neurons which have led to the correct result – it may be green, but it isn't a package – become stronger, while erroneous interpretations weaken the connection. All of this leads to an intelligent neural network changes the connections between the neurons – it “learns”.
The trucks of the future – never again will we have tired drivers in weather-induced chaos
The driver assistance systems of some premium car manufacturers already work in a similar way – as part of these systems, cameras and ultrasound sensors register people and obstacles and make driving safer. However, fully autonomous driving is (still) in the future for passenger cars and trucks. There are too many parameters such as skills, experience and driving behavior of the driver, traffic, light influences, general driving behavior or weather. A Stuttgart-based car firm has announced an AI-based taxi which is due at the beginning of the coming decade.
These vehicles were, however, designed as robot taxis from the very outset – they do not use the technology kit of a production vehicle, instead, they have been developed from scratch. For trucks within the logistics sector, additional prognoses based on technical data such as temperature or oil pressure are conceivable. In the long term, servicing plans would not be aligned so much with manufacturer details – rather, they would be based on the needs and driving behavior of each individual truck.
Detailed evaluation of driving behavior could once again further reduce diesel consumption and CO2 emissions. It is even conceivable that routes which are optimized on a daily basis taking into account traffic and weather conditions could be worked out. Parameters such as travel time, driving time and driver experience could support driver data such as tiredness, to improve staff scheduling.
Avoid empty runs with Deep Learning
In intralogistics too, new prospects are opening up: In the B2B and B2C sectors, handling volumes are increasing continuously. Classic technology has reached its boundaries. Sorting, loading and driving the packages away is taking longer and longer and at the end, it must be compensated for by the suppliers. Manual handling is inefficient, prone to errors and gives rise to costs. Image processing in connection with AI can change that. Deep Learning could take over picking and improve it further – both with regard to recognizing damage and a final quality control. Deep Learning can also deliver valuable, data-based prognoses concerning the increase or flattening-out of demand for goods and therefore help to avoid warehouse bottlenecks or empty runs.
The use of neural networks can shorten itineraries in the logistics centers and optimize picking routes. It is also possible that drones may be used as a method of choice for extensive and challenging inventories of warehouseand goods stocks. The advantages would be huge: the strain would be taken off employees and the potential for danger caused by long periods spent in unhealthy or dangerous surroundings could be reduced. Armbands worn by the employees could record their specific patterns of movement during work in order to make ergonomic suggestions as a result of the evaluation e.g. for a healthy back. Illness-related malfunctions in logistics would be reduced.
Swarms and multi-agent software
It's a fact: the future of logistics, too, is increasingly shaped by Artificial Intelligence and collaborations between humans and machines. Current research and payable innovations in the field of sensor technology, intelligent systems and actuator technology make it possible. The use of driverless transport systems is a first, simple example: in limited areas such as hangars and harbors, these vehicles, which act according to pre-programmed movement patterns, are fully autonomous.
At the present time, the clear delimitation remains between human and artificial decisions. The machine proposes a course of action and the person approves it if they agree. The more intelligent systems become and the closer we get to so-called super-intelligences, the more these boundaries vanish. An experiment conducted at the Fraunhofer-Institut for Material Flows and Logistics (IML) makes this clear: There, more than 50 intelligent transportation vehicles which are also networked with one another interact. The automated assistants transported loads from a high-bay warehouse to workstations in a hangar. They coordinated themselves according to the principle of the Internet of Things by using multi-agent software, both on an independent and decentralized basis. The result: the behavior of the transport vehicles was no longer foreseeable for the humans. The swarm increased or decreased depending on the workload and tasks. All that is not yet currently in the mainstream. Humans need time to be able to trust new, transparent technology. And still today, precisely this level of trust is what sets it apart from Artificial Intelligence.