The Physical AI Revolution in Robotics

SoftServe and Wandelbots are changing how manufacturers use NVIDIA simulation technologies to embed flexibility and intelligence into robotic automation

Advances in automation have long driven manufacturing efficiency; the leap into Industry 4.0 and 5.0 has proved to be no exception. Particularly given the skilled labor shortages and volatile supply chains hampering Western industrial nations, doubling down on automation will be key to remaining competitive. However, with a new global average robot density of 162 units per 10,000 employees in 2023 — twice the number from only seven years ago — the traditional approach to automation innovation will not be up to the task.

Inflexible and expensive, standard robotic automation systems require significant downtime and costly manual reprogramming for every new task. This rigidity makes traditional robots impractical — particularly for small- and medium-sized companies (SMEs) lacking the resources of their multinational enterprise competitors. Without a new approach, competitive dynamics could result in downsizing or even a wave of insolvency.

We see the answer to these challenges in robotics programmed using physical AI, where artificial intelligence bridges the gap between the virtual and the physical world — empowering robots to learn, adapt, and operate intelligently in dynamic environments. Keep reading to explore the physical AI revolution within robotics programming and learn about its context, technical foundations, and promise to transform manufacturing practices.

The challenges facing Western industry: reshoring and skill shortages

The calls for advancements in industrial automation have grown louder as the crises facing Western manufacturers have compounded. Supply chains remain stressed in the aftermath of the pandemic and ongoing geopolitical conflict. To wit: German car manufacturers, long considered leaders in global automotive innovation, face production delays and price hikes due to reliance on overseas suppliers and consequently saw their margins suffer.

Meanwhile, industries that outsourced major segments of their supply chains are feeling intense pressure to reshore. U.S.-based companies like Intel are investing in automated chip factories to reduce dependency on suppliers in the Far East. Yet Western industries continue to grapple with a prolonged skilled labor shortage — hindering their ability to ramp up production. Only a new approach to automation can compensate for the dearth of skilled workers and reliable supply chains.

The need for smart and versatile robots

The standard approach to industrial robots prevalent in the West — often programmed with static instructions — is ill-suited to meet new, flexible production demands. While adaptative robotics alleviated some of the most arduous pain points, setting up or changing their programming requires extensive expertise, financial investments, and time.

Testing robots in the real world is furthermore slow, risky, and expensive — every mistake can damage equipment or disrupt production. These limitations lead to soaring costs for reprogramming and testing every new task, making the technology inaccessible while further exposing vulnerabilities in complex, evolving production processes.

These limitations are largely due to the reliance on an outdated technology stack, which includes proprietary programming languages and control concepts tailored to each robot manufacturer. Additionally, the lack of standardized interfaces for data access and control, coupled with the absence of integration in modern development tools and frameworks, further exacerbates these technological hurdles.

Given these pitfalls, the new industrial landscape demands a new approach to robotics. Adaptive robots need to be capable of seamlessly switching between tasks without costly reprogramming, revolutionizing factory operations for businesses of all sizes. To address these issues, modern manufacturing must deploy a new method to commission intelligent robotics systems, capable of understanding and adapting to diverse tasks autonomously.

Physical AI: the key to a new generation of robotic automation

The bedrock for this new type of automation lies in physical AI. The advancements in this technology are more than a step forward for automation; they represent a fundamental shift in how machines integrate intelligence and adaptability. By merging AI technologies with physical robotic systems, physical AI enables robots to perceive, think, and interact with their environments dynamically.

At its core, physical AI bridges the gap between the data-centric technologies associated with artificial intelligence and the necessities entailed in leveraging physical robots in real production environments. Its benefits encompass more than mere cost reductions during programming. Rather, it introduces a paradigm of flexibility, intelligence, and democratization that fundamentally changes how businesses approach automation.

Scalability and flexibility:
Physical AI enables manufacturers to deploy robotics that adjust to evolving production needs and design changes. For example, a mid-sized automotive supplier can train robots in virtual simulations to adapt to new workflows or product iterations with minimal downtime and no extensive hardware modifications.

Enhanced adaptability and decision-making:
Physical AI empowers robots to handle unpredictable conditions and perform intricate tasks effectively. For instance, a robotic system handling box stacking can adjust to varying box sizes and configurations through AI-based learning, eliminating prior constraints seen in traditional systems.

Cost-effectiveness and faster deployment:
Simulation-based training and testing ensure robots can validate operations virtually before being introduced to physical production lines. This reduces liability, accelerates implementation timelines, and significantly lowers costs by proving feasibility without initial investments into hardware.

From data points to the shopfloor: the physical AI leap

Physical AI was long restricted to highly specialized applications. Its expansion has been powered by advancements in computational power and simulation tools that significantly lower the threshold for entry. Moreover, specific innovations have made physical AI viable for industrial robotics. Specifically, lower thresholds to harness three core components are driving its growth in industrial applications:

Reinforcement learning:
Reinforcement learning uses trial and error at a high scale and pace for skill acquisition. Robots are trained with artificial incentives to optimize performance. For example, robotic arms leveraging reinforcement learning can determine the most efficient way to grip irregularly shaped objects without periodic human interference.

Large language models (LLMs):
The rise of LLMs has introduced the use of natural language in robotics programming. Robots can process commands to generate control behavior and adapt appropriately. That removes the need for specialized robotics engineers for simple yet effective tasks.

Physically accurate simulation:
Lifelike simulation tools are instrumental in the deployment of physical AI. Platforms such as NVIDIA Omniverse and NVIDIA Isaac Sim allow manufacturers to create realistic digital twins of production environments. These simulations prepare robotic programming for different setups — prior to physical commissioning. In existing installations, AI-based path generation optimizes robots’ movements, e.g., to increase output by decreasing cycle times. Simulated ideal paths can be transferred to the physical cell.

Physical AI in action with SoftServe, Wandelbots, and NVIDIA

For manufacturers seeking to navigate the complexities of reshoring and labor shortages with physical AI, a synergy between Wandelbots, SoftServe, and NVIDIA is setting a new standard. The convergence not only simplifies the robotics programming interface but also enhances operational readiness, ensuring continual alignment between simulation and real-world outcomes — at scale across production facilities.

Wandelbots has developed NOVA, a user-friendly low-code software platform designed for easy robotic programming and making robots intelligent, software-driven assets. When combined with NVIDIA Isaac Sim’s physically accurate simulations, NOVA allows businesses to bring robots from virtual space to the real production floor seamlessly. Whether for designing a new production line (greenfield) or upgrading an existing one (brownfield), NOVA enables a seamless transition between simulation and real-world deployment.

Case study in digital twin and reinforcement learning: Volkswagen

To illustrate the potential for a physical AI-grounded approach, consider Volkswagen's project to update the assembly process for roof liner stands.

The production process for these components is notoriously complex due to the sheer size and flexibility of the workpieces, in addition to the challenges of automating the assembly. Crucially, designing robotic grippers to manipulate the components is not only cumbersome but also a significant investment in time and resources. Previously, manually designing automation solutions for such intricate tasks could take years, given the vast scope for solutions and prolonged iteration cycles needed.

Using NOVA and NVIDIA Omniverse, Volkswagen was able to leverage reinforcement learning on a digital twin for its design journey. That allowed it to:

• Utilize randomization to explore multiple layouts and gripper designs.
• Specify material properties to align with production requirements.
• Generate multiple strategies for the assembly process, breaking it into manageable sub-procedures.
• Leverage a GPU cluster to test various variants in parallel.
• Optimize for collision-free operations, minimal cycle times, and stable sub-processes.
• Implement adaptive processes utilizing vision and force sensors with real-time adjustments.

Volkswagen was able to derive multiple working strategies automatically in just 17 hours — a process that traditionally took months or years. That success highlights a pivotal shift towards efficiency and innovation in automotive assembly, setting a benchmark for future automation processes.

A new age of automation — what’s possible when robots think and adapt?

Physical AI is rewriting the future of robotics programming. Its scalable and adaptable nature enables manufacturers of all sizes to reap the benefits of intelligent automation. By reducing costs, increasing efficiency, and eliminating barriers for SMEs, physical AI presents a new chapter for global manufacturing.

Wandelbots’ integration with NVIDIA Omniverse and NVIDIA Isaac Sim ensures that simulation and physical execution stay in sync, enabling real-time testing, validation, and troubleshooting. Whether deployed via cloud or on-premises, SoftServe can help companies use Wandelbots NOVA to reduce development time, lower costs, and confidently scale AI-driven robotics across their operations.

What could your production line achieve if your robots could learn, adapt, and make decisions autonomously? Stay tuned for future articles diving into specific use cases, technical details, and more. Or visit us at automatica in Munich from 24-27 June to talk to our experts personally.