Robots, agents, and copilots unite to turn intelligence into action. By Boris Dmitriev and Daniel Krampe
One of the newest additions to the production floor at Martur Fompak International’s automotive interior component plant in Turkey arrived without a résumé or any prior work experience, but with a well-defined job description: execute a series of repetitive and physically demanding tasks to free others in the workforce to focus on safer, less monotonous and higher-value work.
After just a few months on the job, the new arrival, HMND 01 Alpha, a robot built by the UK company Humanoid, is accomplishing all that and more. Not only is it increasing efficiency across the manufacturing plant and warehouse, it also is enabling employees to focus on inventory management, quality checks, and resolving operational exceptions – tasks that require human judgment. Ultimately, the robot is proving that digital intelligence can be transformed into physical action on the shop floor and in the warehouse with a combination of generative AI, agentic AI and physical (or embodied) AI.
It starts with genAI interpreting material requirements in a timely manner and activating a dedicated AI agent that then orchestrates HMND 01 Alpha (along with other autonomous mobile robots (AMRs) to collect, transport, and deliver materials to production lines. Along the way, agents inform and guide the process with automated prerequisite checks for releasing production orders, including material, capacity, and scheduling availability. They flag material shortages and suggest workarounds, including alternative components and scheduling adjustments.
This is exactly the type of use case manufacturers should be pursuing to maximize their AI investments, according to a recent report from Deloitte, the findings of which are based on a survey of 140+ manufacturers. The report asserts that it’s time for manufacturers to prioritize ‘combining AI technologies deliberately,’ adding, ‘The question is no longer who has the most AI pilots. The question is who can replicate successful use cases across lines, plants, and regions with consistent performance, governance, and user adoption.’
Early results at Martur Fompak support such a strategy. They show increased throughput, fewer errors and an AI-driven intralogistics model that appears readily scalable. The company is now targeting up to five times reduction in manual logistics coordination in the future state.
The business case for multi-AI deployments
Martur Fompak is among a range of manufacturers to successfully pilot multi-AI use cases. Here are two more examples:
- In Vodafone Germany’s warehouse in Duisburg, humanoid robots autonomously executed various visual inspection tasks across the facility. They detected misplaced or damaged products, assessed pallet stacking and weight distribution, highlighted unused storage space, and identified potential hazards such as obstacles in aisles or misaligned pallets. They then reported their findings and recommendations directly into a warehouse system for real-time visibility and better-informed, timelier decision-making.
- Mahindra & Mahindra, a global automotive and industrial manufacturer, transformed an error-prone manual vehicle identity inspection process into a fast, scalable, and automated operation, using agentic and embodied AI with physical AI models to verify cars in seconds (via VIN plate, context engine number, chassis number, etc.) and update systems in real time. The result: an almost 67 percent reduction in manual verification effort and significant time savings.
Initiatives like these are ongoing on production floors and in warehouses across the manufacturing landscape. With its ability to make autonomous, contextual decisions that factor in a broader business conditions (supply chain, demand, etc.), we’re finding embodied AI (that’s physical AI + AI agents, see sidebar) to be particularly well suited for tasks involving a degree of unpredictability (in the task itself or in the task environment), including warehouse pick-and-place, asset inspection, health and safety inspection, quality inspection, and material handling and assembly. It’s also proving to be viable for shop floor orchestration and execution.
To make these kinds of multi-AI use cases successful and scalable, manufacturers need to have a few foundational elements in place. That includes solid, trusted and fresh data from internal sources across all operations and from external sources such as suppliers, as well as a platform that has standard, user-friendly interfaces for running autonomous workflows, with the flexibility to design and deploy customer-specific requirements and workflows. Governance is also critical. The workings and output of AI models must be transparent, easily validated and readily auditable by human beings. While the goal with AI is to let it handle certain tasks autonomously so people can focus on higher-value pursuits, human beings must be able to verify AI is working as intended and hold ultimate responsibility for confirming it’s making on-target business decisions. Training or upskilling employees to execute these human-in-the-loop responsibilities is also essential to any AI deployment.
With each successful proof of concept, the business case for pursuing multi-AI use cases strengthens. The race is on to scale these successes across the manufacturing enterprise, so they become a true competitive difference-maker.
Boris Dmitriev | Daniel Krampe
www.sap.com/industries/industrial-manufacturing.html
Boris Dmitriev is a global industry advisor for Industrial Manufacturing at SAP. Connect with him on LinkedIn at www.linkedin.com/in/borisdmitriev. Daniel Krampe is an industrial manufacturing solution expert at SAP. Connect with him on LinkedIn at www.linkedin.com/in/danielkra. As a global leader in enterprise applications and business AI, SAP stands at the nexus of business and technology. For over 50 years, organizations have trusted SAP to bring out their best by uniting business-critical operations spanning finance, procurement, HR, supply chain and customer experience
