Authors: Bojan Stojkovski and Bogdan Iordache

What does the future of robotics look like through the eyes of someone who’s been driving it forward? For Ralf Gulde, co-founder and CEO of Sereact, the future isn’t just about building robots – but creating technology that can adapt, learn, and make a real difference in how we work. 

Ralf, who describes himself as a “roboticist with a strong focus on machine learning,” began his journey with a deep belief that AI-powered robotics could fundamentally transform industries like warehousing and manufacturing. Robotics startup Sereact is developing robots that don’t just perform tasks but respond intelligently to their environments.

His commitment to AI-driven flexibility has been a key force behind Germany-based Sereact’s success, resulting in a €25 million Creandum-backed Series A funding round that is looking to elevate the company’s growth and expand to global markets such as the US. 

In this interview with Underline Ventures, Ralf talks about his approach to robotics, the importance of adaptability, and how the real challenge lies in making robots not just tools, but also collaborators that adjust to new environments.

UV: Why AI-driven robotics and how does Sereact fit into shaping the next industrial revolution?

Ralf Gulde: Many still perceive robotics as a tool primarily for solving mass production challenges. For decades, robotics has been successfully used in industries like automotive manufacturing, where high-volume production requires robots to perform the same precise, hardcoded movements repeatedly, often with little to no sensor integration. If something in the process didn’t work—such as a missing or misaligned large part—these early robots couldn’t adapt. This was Robotics 1.0.

Today, however, industries with a strong need for automation—such as warehousing and intralogistics—face challenges that Robotics 1.0 couldn’t solve. Imagine a warehouse with hundreds of millions of unique items. Teaching a robot to handle each object, or adapting to slight variations in tasks within such a dynamic environment, proved unfeasible with traditional approaches. There is now a clear demand for more flexible and intelligent robotics.

We’ve all witnessed the transformative power of novel AI architectures, like transformers, which have revolutionized digital productivity—enhancing text generation, translation, and coding. However, productivity gains in the physical world have lagged.

At Sereact, we were among the first to apply these new AI paradigms—such as large language and vision models—to robotics, creating what we call large vision-action models. This gives robots an unprecedented level of generalization, enabling them to operate in complex environments like dynamic warehouses and diverse manufacturing scenarios. Unlike traditional robots, which execute rigid, pre-programmed motions, these AI-driven systems can adapt to different surroundings, objects, and tasks, making robotics significantly more flexible and capable than ever before.

UV: You’re building robots for common use cases in warehousing and industry—how did you come up with this approach, and how does this work in practice?

RG: For our product assessment, it was clear that our target market needed to have a significant pain point and the capital to invest in a solution.

Task complexity was another key factor. We identified intralogistics, specifically pick-and-place tasks in warehouses, as a high-pain-point problem. Companies struggle to find workers for these monotonous, non-ergonomic tasks, and their economic viability depends on efficient order fulfillment. For example, if Zalando cannot ship parcels on time, costs rise, customers become dissatisfied, and they may stop ordering.

Pick-and-place in warehouses is also a relatively structured problem compared to home robotics, where environments are highly dynamic and safety concerns are more complex due to human interaction. In warehousing, robots move items from a dedicated source location to a designated target—whether an automated storage system, a mobile robot, or directly into a parcel.

While the challenge comes from the sheer variety of products warehouses handle, this use case is one of the easiest entry points for AI-driven robotics. It represents the “low-hanging fruit” in the AI robotics landscape, requiring Robotics 2.0 to solve this problem effectively.

UV: Why couldn’t existing robotics suppliers in warehousing, such as those providing robots for picking and packaging, develop this technology themselves?

RG: This is a really interesting question, and it applies to many other industries as well. For example, why do traditional automakers struggle to transition into AI-first software companies? Similarly, most robotics manufacturers come from a mechanical engineering background. While they have strong engineering expertise, many still rely on outdated programming paradigms.

Most robot manufacturers haven’t adopted modern software approaches—there are no apps and even widely used languages like Python are absent from their systems. For a long time, they believed that using proprietary programming languages would protect their market by locking in users. However, this has now created a problem: programming robots isn’t appealing or user-friendly. Instead, it remains stuck in outdated, cumbersome programming styles from the 1970s and 1980s—approaches that no one should be using anymore.

UV: What are the key challenges in developing AI that can operate across different robots and tasks?

RG: Many companies make the mistake of using different models for different tasks or customers, but at Sereact, we follow a “one model” approach. We use a single model for every customer and task, which is crucial because without this approach, maintaining multiple models becomes a nightmare.

What sets Sereact apart is that our Vision Language Action Model (VLAM) generalizes not only across different tasks, like picking items, but also across various robot hardware. Our model abstracts actions mathematically, allowing it to work on articulated robot arms, humanoids, and mobile manipulators, regardless of brand. This flexibility makes our solution incredibly powerful and enables us to provide a layer on top of off-the-shelf robots. Instead of waiting for humanoid robots to become reliable and cost-effective in 5 to 10 years, we can automate today.

This approach also ties into our data strategy. We have systems, like civic pick-and-place robots and direct lenses, that collect hundreds of gigabytes of domain-specific data every day. This real-world data is what strengthens Sereact’s model.

UV: What do you see as your biggest competitive advantage? Can you provide some real-world examples of how your customers have benefited from your solution?

RG: We’ve invested significant research into our VLAM, and we released our first version in early 2023, just around the same time ChatGPT was launched. Our model outperforms some of the major players, including OpenAI, in areas like scene understanding and reasoning. Of course, the real strength lies in the data we’ve accumulated.

With our technology, customers can now change a robot’s behavior using simple natural language prompts. For example, a customer can tell the robot, “If you find trash in the bin, please remove it.” This capability eliminates the need for lengthy development projects with suppliers and allows clients to quickly and easily adjust the robot’s behavior on their own. This flexibility makes robotics far more accessible and useful, transforming them into a tool that can boost productivity—something we desperately need in today’s environment.

UV: How do you build trust with traditional industrial customers and overcome the pushback you encounter from them?

RG: As a young B2B company, the most important thing we have is the trust of our clients and our reputation. We prioritize delivering on our promises and ensuring our clients are exceptionally satisfied. This focus on client satisfaction was also a key reason why Creandum was so interested in Sereact—they spoke to every customer of ours and found that we exceeded their expectations. This is a lot of hard work, and it will continue to be a major challenge for Sereact, but I make it the number one KPI for our team.

However, when selling to major automotive companies like Mercedes-Benz, questions inevitably arise. They might ask, “Can you provide the same level of support as traditional manufacturing companies or old-school system integrators?” For example, “Can you deliver top-notch performance if the system breaks down in the U.S. at 2 a.m.?” And of course, we can’t make those promises. That’s why we partner with companies that specialize in integrations and can handle that support. We bring superior software to the table, which helps us scale like a software service, while system integrators now have access to new markets.

For example, we’ve recently partnered with AWL, one of the leading robot integrators, known for their expertise in setting up robot cell safety technologies. By combining their strengths with Sereact’s software, we’ve created a collaborative product that delivers excellent performance for our end customers.

UV: What is your take on choosing a software-first approach versus hardware-focused solutions?

RG: Our passion is AI software, and our approach is fully software-defined, allowing everything to be configured through software during deployment. This makes our solution highly flexible and adaptive. However, we’re not afraid to work with hardware when necessary. For instance, we developed and patented our own gripper, which you can find on select AI platforms because we saw that the right tool was missing. 

Without it, performance wouldn’t be as strong as it could be. While we do innovate in hardware when needed, we would never try to build a robot itself, as the robot market is already saturated with great products. We focus on filling the gaps where necessary.

UV: You recently closed a €25 million Series A round. What are your plans moving forward with this funding?

RG: We only decided to raise funds because we saw a clear opportunity. Sereact has always been capital efficient—my co-founder and I were the ones buying old gamer GPUs and building our servers, always looking for maximum capital efficiency. However, we reached a point where we felt we needed to scale up. While we were making good progress and growing, both Marc (Tuscher, co-founder of Sereact) and I felt like we were waking up with one foot on the brakes every day. We knew we had to double down on compute power to train larger models capable of handling more complex manipulations, which would allow us to enter more advanced markets like manufacturing and return handling.

I also strongly believe that Europe is the best place to start a robotics company due to its manufacturing base and the infrastructure needed for robotics—like conveyors and automated storage and retrieval systems. Europe has all the foundational elements to create fully automated warehouses, making it the ideal starting point. But as we saw year-over-year growth in the US, we realized the need to expand there as well.

So, we decided to fundraise. The whole process happened very quickly, within just a few weeks, and we were thrilled with the outcome. The round was highly competitive, and we were very happy with Creandum. One thing I noticed is that building Sereact is tough—our team is deeply committed to delivering best-in-class results. It reminded me of what I saw with the team at Creandum. That level of dedication is exactly what Marc and I saw when we were in Silicon Valley meeting investors, and it’s the kind of passion we want from our partners.

UV: How competitive do you find the US market industrial tech-wise?

RG: In the US, I don’t see any solution like Sereact right now. There was Covariant, which was acquired by Amazon, and now there’s a gap, which makes the opportunity even more exciting. Of course, there are cultural differences between Europe and the US when it comes to adopting automation in warehouses, but I believe our solution will resonate well. We’re already seeing strong demand for parcel handling, but not the complex commissioning that Sereact specializes in. 

UV: How do you see humans and robots working together in the future?

RG: Robots augmenting humans and taking on dangerous tasks is a great shift. When I was sixteen, I did pick-and-place work in a warehouse, and my biggest challenge was a huge clock above me that didn’t move, highlighting how boring the task was. 

That’s why I believe it’s fantastic that robots can handle these dangerous and monotonous jobs, leaving humans to focus on more engaging work. With demographic shifts, we’re seeing a shortage of people willing to do such tasks. That is a major issue. I believe that, very soon, every human will have a robotic system, and I’m excited for that future.