In the foreseeable future, technology will no longer be an obstacle to achieving fully automated factories with minimal human involvement. Instead, the decision to pursue lights-out transformations will largely hinge on economic considerations. Manufacturers that embrace automation and demonstrate agility in overhauling their operations will be best positioned to benefit from this shift.

For decades, the manufacturing sector has anticipated the rise of fully automated factories. In these environments, production would be controlled by networks of high-tech robots, intelligent machines, and sensors, addressing widespread labour shortages while significantly cutting operating costs. With minimal human intervention, these factories could theoretically operate in complete darkness, earning the term “lights-out factory.”

However, notable attempts to create such factories, like Adidas’ Speed Factories in the U.S. and Germany, and Stanley Black & Decker’s Texas-based facility, have raised doubts about the viability of this concept. Additionally, Tesla’s automation efforts have further contributed to these concerns. Even Elon Musk acknowledged that “excessive automation… was a mistake”. This led some experts to recommend abandoning the idea of lights-out factories altogether.

A group of BCG consultants argues that this would be a mistake. The near-stagnant growth in manufacturing output per hour worked in mature economies, such as the U.S. (-0.4%) and Germany (1%) since 2018, underscores the need for automation. Moreover, this stagnation highlights the importance of driving productivity gains through increased automation. The good news for manufacturers is that, based on the group’s research and practical experience, a significant shift is underway. The barriers that hindered earlier attempts at automation are set to rapidly diminish in the coming years. Robots are becoming more capable, flexible, and cost-effective, with AI bringing new levels of intelligence into the factory environment. Manufacturers must prepare for this inevitable disruption or risk falling behind.

 

Transforming an Existing Factory to Lights-Out

Globally, some manufacturers are already proving the long-term viability of lights-out factories. For example, one of the group’s clients, a Europe-based automotive supplier, made a bold decision to transform its existing setup into a fully automated operation. In this new model, humans serve as planners, supervisors, and maintainers, rather than performing the physical tasks.

The company aimed to achieve cost-competitiveness, making it economically viable to operate in a high-cost country. Additionally, they focused on continuing to deliver to customers within two days. By reducing direct labour by 100%, the company effectively addressed the talent shortages that have plagued the industry. The factory also saw an 8% improvement in EBITDA. This successful transformation is now being scaled across multiple high-cost sites globally.

To initiate this transformation, the client proactively addressed the most common financial and technical challenges faced by lights-out manufacturers. These challenges are found in all categories of factory operations.

 

Category 1: Predictable Processes

Most factory processes involve repetitive tasks, making them technically feasible and profitable for robots to perform. For instance, self-guided vehicles can easily navigate the shop floor along predetermined paths. However, installing and maintaining these robots requires a highly skilled workforce, which limits the potential for overall workforce cost reduction.

 

Category 2: Unpredictable Processes

Even in tightly controlled environments, some factory processes remain inherently unstructured. Pre-programmed robots often struggle to adapt to complex, unexpected situations. For example, automated quality control systems may fail to identify defects that occur for the first time. This lack of adaptability prevents manufacturers from fully trusting automation.

 

Category 3: Non-Automatable Processes

Current technology limitations mean that certain intricate processes cannot be automated. For example, robots struggle with “kitting,” the process of picking and organising various small parts for assembly. These parts are often too similar for robots to identify easily or too different for them to grasp correctly. As a result, human workers are still needed to complete these tasks.

 

To overcome these challenges, the client adopted a “redesign for automation” approach for its processes, products, and layout. This broad overhaul of factory operations added new process steps to improve automation feasibility while removing human-oriented inefficiencies. For example, the client no longer needed to sacrifice floor space for storing inventory that humans could see and reach. Instead, they built vertical storage areas that robots could easily access and navigate. The freed-up space allowed them to install more machines, increasing output by more than 30%. By essentially rebuilding processes from the ground up, the client uncovered innovative adaptations.

 

Adaptation 1: Optimising Robot Use to Minimise Costs

During the transformation, the focus was on streamlining the number of robots deployed. For example, the client faced a key decision early on: whether to invest half a million dollars just to transport plastic moulded parts to the assembly station. This expense was high because each part size required a specialised robot. To avoid this cost, the consultants helped redesign the process. They added a conveyor belt that grouped the outputs of similarly sized plastic parts together. This single new step reduced the number of required robots by half, significantly improving the business case for the lights-out factory.

 

Adaptation 2: Learning to Anticipate Future Scenarios

The objective was to train the AI systems to prepare for uncertainty before production began, rather than during it. For example, the client installed an automated visual inspector to detect quality issues such as scratches. Some defects occur so rarely that it can take years for the system to learn how to detect them without human support. To address this, the consultants generated a large volume of artificial or “synthetic” images of potential defects, simulating various lighting conditions. This additional step allowed the deployment of a fully trained visual inspector from day one. It this, eliminating the need for supervision until the system stabilises.

 

Adaptation 3: Adjusting the Process to Avoid Certain Tasks

Process flows can be modified to bypass tasks that cannot yet be automated. In the client’s case, no single robot could effectively perform kitting. To phase out this task, the consultants established a QR-labelled box warehousing system in collaboration with suppliers. Each box contained a single part type. This enables robots to autonomously identify QR codes and retrieve boxes in the exact sequence required for assembly.

 

The client’s lights-out factory demonstrates the tremendous potential of this approach. However, the transition has not been easy for all manufacturers. Many perceive the technical barriers as too prohibitive and are hesitant to make the necessary investments. They fear disruption to their entire factory operations. Nonetheless, this mindset is set to change drastically, making the concept of lights-out factories more accessible to a broader range of manufacturers.

 

AI Is Lowering the Entry Barriers for Lights-Out Factories

We are now entering a period of transition that is bringing down the entry barriers for manufacturers of all sizes. Lights-out automation depends on robots that can perceive, plan, and perform tasks at nearly the same level as humans. Until recently, these robots only supported pre-planned, single-dimensional interactions with the factory environment, severely limiting their capabilities.

This is changing as embodied agents bring the power of Generative AI into the physical world of robotics. Now, robots can perceive changes on the shop floor through multimodal senses, including text, audio, sensors, and signals. AI can then generate simple plans and iterate on them to suit this understanding of the environment. In real-time, robots can translate these plans into specific actions and execute them. Although these developments are still new, many have already entered the R&D pipelines of manufacturers and are beginning to be applied.

 

1. Instruction Simplicity

Programming and integration account for 50% to 70% of the cost of a robotic application. Generative AI interfaces are expected to significantly reduce this cost by providing a natural language interface that allows even non-technical workers to instruct robots. The transformation would be drastic. Instead of one specialised worker for every eight robots, the factory would require one non-specialised worker for every 25 robots.

Industry applications have already emerged. For example, Sereact has rolled out a voice or text command interface named PickGPT. This interface allows users to interact with robots using simple instructions like “I need to pack the order.”

 

2. Task Versatility

The traditional approach of training robots for single tasks is becoming obsolete. Recent research shows that generalist robotic AI outperforms specialist robots by more than 50% across diverse tasks. Such tasks include cable routing, object manipulation, and pick-and-place. This versatility reduces the need to deploy numerous robots, minimises downtime, and enhances overall effectiveness.

Currently, auto manufacturers like BMW, Mercedes, Honda, Hyundai, and Tesla are all exploring task versatility through general-purpose humanoids. For example, Figure AI, backed by OpenAI, is being piloted at BMW across its body shop, sheet metal, and warehouse operations. However, in lights-out environments, where human-robot interaction is minimal, the humanoid appearance is unnecessary. Therefore, the technology is shifting towards more economical designs while maintaining the same versatility. For example, Mimic is building humanoid hands without legs.

 

3. Situation Adaptability

Robots can now dynamically identify and adapt to changing industrial environments. For example, Covariant’s robotic foundation model devises its own strategies to handle challenging pick-and-place tasks, such as shaking boxes to grasp items more easily. This capability allows robots to manage over 100,000 previously unseen SKUs in a customer warehouse from day one. Covariant’s models can adapt well because they have learned from millions of real-world multimodal datasets of robots picking objects. Pre-trained models will level the playing field for manufacturers by cutting setup time and boosting efficiency with on-the-fly adaptation. Additionally, they will build trust in lights-out setups.

 

4. Human-Like Dexterity

The range of automatable tasks is rapidly expanding as robots improve their ability to mimic human behaviour. The Toyota Research Institute is developing a “kindergarten for robots,” where human tele-operators manually control robots to perform over 200 common human skills. These skills include tasks such as using tools and pouring liquids. Meanwhile, a Generative AI model learns these behaviours in the background. Google has demonstrated that robots can even learn skills from videos by mapping out human movement trajectories. Just in the last year, automation solutions for the long-standing challenge of kitting are finally becoming industrialised. Daimler is deploying robotic arms with fine dexterity to handle electric vehicle kitting across its battery production lines.

 

We are also starting to see the advent of factory planning software that incorporates digital twin technologies. This software allows operators to visualise the impact of a change in seconds rather than months. Virtual factory tools, like Cosmo Tech’s Digital Twin Simulation, are also expected to help designers assess how robotic AI would respond to a future, as-yet-unseen production problem. This capability will give manufacturers the confidence to adopt lights-out operations.

 

Where Do Manufacturers Go from Here?

What does this all mean for manufacturing? The consultants’ message is clear. In many cases, the only way to significantly improve manufacturing productivity will be to move to a fully automated lights-out factory. This means that manufacturing leaders who are not actively testing their next big AI investment are already behind.

There will be resistance to automation from many quarters. Some companies may not find the economics compelling, while others may prefer to wait for further proof of technology maturity. Yet, manufacturers cannot afford to stand still while their competitors take the plunge into automation.

The opportunity for automated factories is here and now. Businesses that delay may soon find themselves falling behind. Investing in AI and automation today will help forward-thinking manufacturers build the lights-out factories of tomorrow.

 

 

---

 

At AFS, we are passionate about fostering innovation and empowering ambitious minds to flourish. Our mission is to provide best-in-class financial services for traditional and crypto deals, exploit European grants, and use quantitative methods to improve clients’ performance. We aim to help our customers unlock their full business potential.

Let’s unlock your enterprise’s full potential together!

Get in touch at [email protected].