Historically, automation required specialized skills and resources, limiting its adoption. Expertise in computer-aided design (CAD), robotics and programming was essential, creating a gap between digital-forward companies and those lagging behind. The digital acceleration of 2020 exacerbated this divide, leaving many manufacturers struggling to keep pace. Organizations’ reluctance to adopt new technology stems from a variety of reasons.
Get your subscription to Control Design’s daily newsletter.
- High perceived cost: The perceived high cost of automation is often cited as a significant barrier to its adoption. However, this perception is more due to a lack of understanding of the direct benefits, rather than the actual expense. To overcome this, it is crucial to illustrate the concrete advantages of automation in terms of productivity and profitability, beyond just the straightforward return on investment (ROI).
- Analysis paralysis: With thousands of autonomous-mobile-robot (AMR), robot and “solution” vendors, organizations often struggle to decide which technologies to implement. The vast array of choices can lead to decision fatigue, where companies delay automation projects due to uncertainty and the fear of making the wrong choice.
- Inefficient processes: Automation applied to an efficient operation will magnify the efficiency. Automation applied to an inefficient operation will magnify the inefficiency. Installing an automation system often requires analyzing your processes, eliminating the waste and making the necessary adjustments for robots and automation to shine. Without the proper visibility and traceability already in place, an automation project can easily transform into a digitalization and process optimization one.
- Sector-specific challenges: Different sectors face unique barriers, from high costs and technology readiness to regulatory policies and customer demand. For example, the healthcare sector must navigate stringent regulatory requirements, while the manufacturing sector may deal with challenges related to technology integration and workforce training.
Can AI be the answer?
Artificial intelligence has undergone significant evolution since its inception. The concept of AI dates back to the mid-20th century when pioneering researchers began exploring the possibility of machines that could mimic human intelligence. Early AI systems were rule-based and limited in scope, capable of performing specific tasks but lacking adaptability.
Over the decades, AI has progressed through various stages of development, from simple algorithms to complex machine-learning models. The advent of deep learning in the 2010s marked a significant milestone, enabling AI systems to learn from vast amounts of data and perform complex tasks with remarkable accuracy. Today, we stand at the cusp of a new era of AI-driven innovation, where generative AI is transforming industries at an unprecedented pace. Generative AI refers to a class of artificial intelligence systems designed to create new content, such as text, images or music, by learning patterns from existing data. Unlike traditional AI, which typically focuses on recognizing and processing data, generative AI can produce original outputs that mimic the style or structure of the input data it was trained on. Large language models (LLMs) are a specific type of generative AI that focuses on understanding and generating human language. Examples include OpenAI's GPT-4 (close source) and Meta’s LLaMA (open source), which are trained on vast text datasets to perform a wide range of language tasks. Foundational models encompass large-scale, pre-trained models like LLMs, serving as a base for various AI applications. These models are characterized by their broad capabilities and ability to generalize across multiple tasks, often being fine-tuned for specific applications to enhance performance.
Generative AI’s potential and challenges
The dream of every robotics user would be a system that can be programmed with natural language commands:
- “pick from this bin”
- “fill this box with six products.”
This system autonomously adapts to changes in its environment and can be debugged and improved with similar natural language commands. Is generative AI getting us closer to that dream? Generative AI could definitely aid in controlling robots, but achieving the precision and speed required in manufacturing remains a challenge.
As the robotics pioneer Rodney Brooks has stated, every successful AI application in production typically possesses two critical characteristics:
- Human in the loop: The world is complex and full of edge cases. Human involvement is crucial for monitoring critical AI systems, intervening when necessary and making nuanced decisions that machines cannot handle alone.
- Low risk if it fails: AI deployments in production often focus on scenarios where failures pose minimal risks. Bin picking is a good example. It doesn’t matter if the robot misses a pick, as long as it can achieve the required cycle time.
When looking into robot control tasks, the speed and precision required make it difficult to adopt AI in the loop. That is why most successful AI applications in manufacturing apply to quality control, material handling and bin picking, where the consequences of a failure are not critical.
Robotics has yet to experience its "ChatGPT moment," largely due to the complexity and variability of physical tasks compared to language tasks. Foundation models like LLMs excel in domains with vast amounts of data and compute power, capturing intricate patterns that enable remarkable predictability, as seen in natural language processing with models like ChatGPT. However, for robotics, the challenge is greater because it involves mastering countless physical simulations that need to generalize to the unpredictable real world. Unlike language, where tasks can be uniformly expressed as text input and output, robotics requires interpreting a wide array of sensory inputs and executing precise physical actions.
Training a "foundation agent" for robotics would necessitate scaling up across numerous simulated and real-world environments, embodying diverse scenarios and tasks. This complexity means that, while the principles of foundational models apply, the path to achieving a universally adaptable robotic agent is more intricate and resource-intensive, delaying its breakthrough moment compared to AI in text-based applications.
Looking into the future, end-to-end learning for robots, in particular multi-purpose ones will see the levels of accuracy needed in manufacturing. There is significant gravity in the research from AI giants like Nvidia, such as project GR00T, and startups like Figure or Apptronik to create embodied agents that learn to perceive and interact with a complex world, bootstrapping intelligence along the way.
However, currently the success of generative AI in manufacturing comes from its role as a co-pilot rather than a pilot. While AI can provide valuable insights, suggestions and automation capabilities, human expertise is indispensable for critical decision-making and overseeing complex processes.
Enhanced decision-making
As engineers, we gravitate toward robot control tasks when thinking about potential impact of generative AI into robotics. However, there are plenty of secondary tasks that don’t require the accuracy of control, where generative AI can act as a multiplier.
AI-driven insights can facilitate better decision-making, accelerating sales and implementation cycles. By analyzing vast amounts of data, AI can provide actionable insights that help companies optimize their operations and improve efficiency. For instance, AI can predict equipment failures, recommend maintenance schedules and optimize supply chain operations, leading to reduced downtime and increased productivity.
The role of generative AI
Robots and generative AI automate tasks, not jobs. The rise of generative AI will inevitably reshape job roles. While it automates repetitive tasks, it also opens up new opportunities. Generative AI can significantly augment various roles and industries:
- Controls engineers: Establish secondary loops to optimize machine operations. AI can help monitor and adjust machine parameters in real-time, improving overall system performance and reliability.
- Mechanical drafters: Create detailed design drawings and modify designs to correct deficiencies. AI can generate detailed drawings and models, freeing up time for innovation and design improvements.
- Robotics technicians: Repair and maintain robotic systems, troubleshoot issues and train personnel. AI can assist in troubleshooting, performing preventive maintenance and training personnel to use and maintain robots. Manuals and code can be queried in real time to help diagnose problems. New insights can be stored without the need of rigid data models, therefore exponentially increasing the amount of information available.
- Sales engineers: Develop tailored proposals, deliver technical presentations and identify resale opportunities. AI can assist in creating proposals and presentations, allowing engineers to focus on more strategic tasks.
- Sales representatives: Prepare sales contracts, respond to inquiries and offer technical support. AI can help negotiate terms, highlight product features based on customer needs and provide ongoing technical support.
- Project management specialists: Delegate tasks, monitor project milestones, and prepare budget estimates. AI tools can handle routine project management tasks, enabling managers to concentrate on critical decision-making.
Empowering less automation-savvy companies
For automation to shine, companies need to focus on optimizing their processes first, and generative AI can help on that task. Generative AI agents can help companies of all sizes diagnose where waste and inefficiencies are, discover which automation solutions have higher potential for impact, rank vendors and proposals, efficiently monitor the purchasing and deployment processes and help with support once the system is installed. Traditionally this would require complex and costly data collection and ingestion into rigid data models and planning systems. In the new world, generative AI enables the digestion of unstructured data, making it easier to analyze and utilize diverse data sources, thus simplifying and accelerating the optimization and automation processes. AI can provide predictive analytics, automate routine tasks and enhance decision-making processes, enabling even less tech-savvy companies to harness the power of automation.
Future outlook
The future of generative AI in manufacturing machinery looks promising as the technology continues to evolve and mature. Advances in AI are expected to improve the accuracy and adaptability of robots, enabling them to perform more complex tasks with greater precision. An increasing number of startups and AI pioneering companies, driven by the humanoid race, are actively working on creating embodied agents that can perceive and interact with the physical world more intelligently. These developments will likely lead to end-to-end learning for multipurpose robots, which could operate in unstructured environments requiring high levels of dexterity and autonomy.
In the near future, generative AI will also democratize access to advanced automation technologies, making them more accessible to smaller companies and allowing them to compete with larger, more established players. As AI technologies continue to integrate into manufacturing, the role of human expertise will remain vital. AI will act as a co-pilot, providing valuable insights and automation capabilities while humans oversee critical decision-making and complex processes. This collaboration between AI and human workers will lead to more efficient and innovative manufacturing environments. Overall, the integration of generative AI into manufacturing machinery is set to transform the industry, driving advancements in automation, improving efficiency and making cutting-edge technologies accessible to a broader range of companies. AI-driven insights will help companies predict equipment failures, recommended maintenance schedules and optimize supply-chain operations, thereby reducing downtime and increasing productivity.
Conclusion
Bridging the automation divide requires innovative solutions and a commitment to making advanced technologies accessible. Generative AI, coupled with strategic data-driven approaches, offers a promising pathway to a more efficient and automated industrial future. By lowering barriers to adoption and providing tools and resources to support companies in their automation journey, AI-driven solutions can help bridge the gap and ensure that all organizations can benefit from the transformative power of automation.
