Robots‌ ‌and‌ ‌AI‌ ‌in‌ ‌the‌ ‌Warehouse‌ ‌

Artificial intelligence (AI), robotics, and warehousing are three of the hottest buzzwords right now. Among all the hype, what is truly materializing in the industry? – A lot! “Dark” warehouses, where the lights can be switched off because robots do all the work, are becoming a reality. This is just the beginning. In the US and Europe, only up to 10% of warehouses have some level of robotic automation, and in developing countries, it is less than 1%. This article will explain the driving forces, the most powerful robots, and the challenges ahead.

Market Forces

More than 4 million robots will be deployed across 50.000+ warehouses by 2025, according to ABI Research, a market research firm. Concurrently, the warehouse automation market is expected to double from 15B USD to 30B USD by 2026.

Three driving forces are behind this sudden rapid adoption of robots and automation equipment. First, robots have become more capable due to sensor improvements (LiDAR) and advances in computer vision (e.g., mapping for mobile robots and pick-and-place for robot arms). Second, the price of robots has halved over the last two decades. Third, the rise of e-commerce,  COVID-19, and labor shortages have forced warehouse owners to look at alternatives for human labor. 

Why is warehousing a hotbed for robotics? 

Warehouses around the world all follow a similar process: goods come in –  goods are stored –  goods move out. As a result, robots that assist in the essential tasks of picking and moving material are relevant to warehouses everywhere, providing economies of scale. 

However, the economic environments in which warehouses operate are wildly different. Some distribution centers are located near expensive urban centers, others in cheaper rural areas. Some have existing buildings and equipment; others will be built from scratch. Some consist of multiple floors; others are housed inside a high-ceiling industrial hangar. 

Besides the building and location, there is variation in the order volume and handling. Some warehouses only store whole pallets; others do individual case picking. Some stock large quantities of just a handful of products; others stock small amounts of a diverse array of products. Some operate 24/7, others only during business hours, and so on.

This diversity in economic environments calls for a wide variety of robotic solutions that are relevant in particular settings. 

Preserving existing infrastructure

Buildings often comprise 20-40% of the total warehouse budget. Making modifications to existing buildings and infrastructure can be costly. For example, good-quality racking is expensive: they consist of tonnes of high-grade steel, and the installation costs should not be underestimated. Warehouse owners would rather not dispose of it before the amortization period. Moreover, a drastic overhaul of the warehouse infrastructure may disrupt current operations. 

One Israeli firm has developed an innovative robot that can operate in most existing warehouses. BionicHIVE’s  “SqUID” robot can move around the warehouse and pick up items from high shelves by trawling across the racks vertically and horizontally. The only requirement is that additional rails click onto the existing frames so the SqUID can grip them. 

Maximizing storage density 

“Stop air housing, start warehousing” – the slogan of Norway’s Autostore – cuts right to the heart of a salient warehousing problem. Around 50% of the warehouse space is occupied by … nothing! Isles are essential for robots and pickers to move around and collect items. But what if goods could move “through” the racks by themselves? – This is precisely what Autostore’s automatic storage and retrieval system (ASRS) has managed to do. 

Designed like a cube, goods are stored at great density inside the Autostore. To get the goods out, a set of trolleys move along rails on top of the cube and “dig up” the desired bins. Although pricey in absolute terms, this solution can be very cost-effective in places with expensive real estate. 

Loading and Unloading 

The focus of some innovative robot manufacturers has shifted to an even more challenging problem than moving material within the warehouse. Goods arrive at the warehouse in trucks of various sizes. Getting the goods out requires maneuvering heavy loads inside of a confined space. Moreover, pallets and boxes often need to be unstacked and restacked to make them ready for put-away. 

If there is one firm that could be up to this challenge, it is Boston Dynamics. The builders of the best-in-class robot dog (Spot) and humanoid robots (Atlas) have recently released a video of their latest family addition: Stretch. 

The most impressive part about Stretch is its lightweight robot arm. To unload boxes faster than humans, a high-speed high-reach robot arm is needed. Typically, these arms are so heavy that they cannot be mounted on a small chassis. Boston Dynamics re-designed the robot arm to be four times lighter than arms with equivalent capabilities, allowing them to mount it on a chassis that fits inside a freight container.   


As with every new technology, solutions create new challenges. In robotics, these challenges revolve around communication and coordination. Each robot manufacturer chooses a different communication protocol (e.g., TCP-IP, MQTT, EtherCat, customized APIs, etc.) to allow the outside world to connect with their system. In other words: robots, barcode scanners, inventory software, doors, elevators, drones, and sensors all speak different languages. 

For the warehouse owner, this presents a challenge. When they want to automate, they suddenly find dozens of devices from different vendors that communicate through various protocols. Making them work together takes weeks and months of bespoke software engineering connecting devices A to B, C to D, E to F, F back to A, etc. 

For example, an autonomous mobile robot (AMR) that moves goods from the inbound area to the shelves needs instructions on where to go from the warehouse management system (WMS). If the warehouse has multiple floors, the robot should be able to take the elevator. When the cleaning robots come in at the end of the day, they should tell the material handling robots to get out of the way. In addition to these very common communication scenarios, there are many edge cases that require communication. 

While bespoke software engineering does the job to connect different systems in a time frame of weeks or months, there is a need to rapidly integrate robots of different brands, make them interoperable, and have the flexibility to add more devices as you go.

Therefore, the future of system integration is no doubt no-code or low-code. Robots and devices will be connected and configured through an app, saving months of software engineering, and making it possible to deploy robots on demand. This technology will allow warehouses to scale up their level of automation along with their order volume.

Coordination and AI-based optimization 

The next step is making fleets of robots coordinate with each other to move efficiently around the warehouse. Firstly, coordination is needed for collision avoidance. Especially on junctions, robots of different brands have trouble figuring out by themselves who should go first. Often, both vehicles will either collide or stay put, waiting for the other to pass first. Another source of collisions is the height of the LiDAR sensor. Many AMRs scan for obstacles using LiDAR in a horizontal 2D plane. When a forklift is parked in an aisle with its forks sticking out above the height of the AMRs LiDAR sensor, the AMR may not perceive the obstacle ahead and run into the forks.

Secondly, coordination is needed for resource optimization: fulfilling as many orders as possible, in the shortest amount of time, with the least amount of movements. An optimal flow depends on many different variables: the layout of the warehouse, the spatial arrangement of the racks, the number and type of robots (“agents” in AI-speak), the order volume, and more. 

In addition, there is the slotting problem. Which items to store on what shelf? Popular items should be kept at a more accessible location than items that are rarely ordered. One more complication is the size of each item and the size of each bin. Big items cannot be stored in small bins. 

Hence, optimizing warehouse operations is a multi-variable, multi-agent, multi-target problem whose parameters change over time. Such intractable problems are traditionally solved with common sense, experience, and rules of thumb. With AI, we have the tools to develop smarter slotting, more efficient layouts, and optimal coordination between different fleets of robots. How is this done? 

In our company, Syncware, we create simulations of the warehouse environment based on real warehouse data, including the floorplan and order history. In the simulation environment, we train the agents (who represent robots or humans) to fulfill orders in the fastest way possible whilst taking into account the movements of the other agents and future incoming orders. On the shop floor, the trained AI informs each robot of their next move.   

As with any AI, the solution is as good as the data. To optimize warehouses not only in theory but also in practice, knowing in real-time which robot is where, how much battery power it still has left, and what orders are expected, is key to giving the best commands to the fleet. 


In the coming years, new compelling robots will be invented, and find their way to the warehouse. Several challenges around connectivity and coordination need to be overcome to orchestrate robots so that managers can truly turn off the lights. However, with a growing global population that is fond of online shopping, the future is certainly bright for dark warehouses. 

Author Profile

Dr. Dirk-Jan van Veen is the co-founder and CEO of Syncware, a start-up specializing in robot connectivity and AI-based warehouse optimization. Before founding Syncware, Dirk-Jan completed a Ph.D. in Multi-Agent Systems at ETH Zurich.


Related post