Centro SIMES Students Achieve Global First in Robotics with NVIDIA’s Isaac GR00T

Santiago, Chile – April 20, 2025 – Just one month after NVIDIA launched its groundbreaking artificial intelligence platform Isaac GR00T, high school students at Centro SIMES in Santiago, Chile, have become the first in the world to use it to train a digital twin of a humanoid robot. The team, aged 15 to 17, is now preparing to transfer this training to real-world operations using a Baxter robot housed at Santiago University.

LLMs (large language models) such as NVIDIA’s ChatGPT have quickly integrated into all sorts of digital platforms and we humans have just as quickly come to rely on, adapt to and greatly trust the algorithms now undergirding daily life–on screens, at least. The scalable leap from AI to humanoid robotics–the evolution from disembodied HAL 9000’s and J.A.R.V.I.S.’s to autonomous C-3POs and WALL-E’s–remains elusive for a myriad of hardware, software and firmware reasons.

Sim-to-Real is the process of training AI in a simulated environment and then transferring the trained model to a real-world robot, where the robot must perform tasks without the controlled conditions of the simulation. This process is not just about running AI models but also adapting them to dynamic, real-world variables such as noise, unexpected obstacles, and real-time decision-making, like a parent observing and correcting every wobbly step of a baby learning to walk before he can successfully navigate the world on his own. Considering both university labs and tech companies full of experienced engineers are challenged by the literal nuts and bolts of AI robotic integration, a small high school academy in Santiago, Chile, is perhaps one of the least likely places for this very early type of scalable integration to happen.

Chile, a country whose name is derived from a Mapuche word likely meaning "where the land ends" or "the deepest point of the earth," embodies the spirit of exploration and discovery that defines Centro SIMES founder and AI robotics team mentor, Claudio Opazo. Driven by a mission to push the boundaries of STEM education, Opazo was inspired to pursue Sim-to-Real AI robotics integration with GR00T after his students successfully qualified for the RoboCup 2025 by leveraging two key algorithms at the heart of GR00T, Eureka and DrEureka.

“Since we'd never participated in RoboCup before, we knew that to qualify, we had to do something truly innovative and high-level. So we used the Eureka and DrEureka algorithms,” Opaza explains. “Eureka and DrEureka automate the creation of reward functions, which are then trained and evaluated in 3D simulations with digital twins. My students used these LLMs to automate and accelerate code creation, enabling them to compete in the 3D soccer simulation league within the RoboCup training environment—without relying on advanced prior knowledge.”

“This approach allowed us to qualify on the first try! We are the only high school team in the world to qualify in that category. All the other teams are students from prestigious universities,” Opazo exclaims, including institutes such as Offenburg University in Germany, China’s Nanjing University and Brazil’s Aeronautics Institute of Technology (ITA).

“After learning that we qualified for the RoboCup on our first try, we set an even more difficult goal: to be the first high school students in the world to use Project GR00T. To achieve this, we've worked very hard to train the Baxter humanoid robot in simulation. And now we will integrate it with a Baxter that they have at a university in Chile.”

Opazo’s students are possibly the first high school students worldwide—and potentially among the first students at any educational level—to engage deeply with NVIDIA’s Isaac GR00T software in humanoid robotics. Remarkably, Centro SIMES’s work with digital twins surpasses even university-level research, including that of institutes equipped with similar robotic technology and high-performance computing resources.

Baxter, from the now defunct Rethink Robotics, is known for its adaptability and human-like dexterity, and is equipped with advanced sensors and vision systems for real-time human collaboration. Wired Magazine eulogized Rethink Robotics by calling Baxter a “charmer” for serious robotics researchers and “legend among engineers in these early days of advanced robotics.” “This robot was once the standard for conducting research in humanoid robotics, which is why it remains quite advanced even today,” Opazo explains, “Using Project GR00T with the Baxter robot is so advanced that not even the university that owns the robot has done this with its university students yet.”

Baxter, as beloved it may be, was introduced in 2012 and contains systems designed for a different era. Modern AI systems like Isaac GR00T demand much higher computational power and more advanced sensory processing capabilities. Integrating these new AI models into Baxter’s older hardware requires overcoming significant technical challenges. Teaching the old Baxter dog new AI tricks, requires Opazo’s young team to not only train a modern LLM, but also construct bridges to systems created more than a decade ago. Major stumbling blocks include:

Hardware Limitations: Many robotic systems, including Baxter, were built with limited computational power, requiring workarounds to handle advanced AI models, especially those as complex as GR00T.

Real-Time Processing: AI models need to process and respond in real-time to sensory input (vision, touch, etc.), which requires robust hardware and extremely efficient algorithms.

Firmware and Compatibility: Adapting firmware to handle modern AI models like Isaac GR00T requires bridging gaps between much older robotics systems and the latest AI advancements.

Sim-to-Real is more than training models–it’s ensuring those models can respond seamlessly to a constantly changing environment. It isn't just about creating models that work in simulations but also about ensuring that these models can handle real-time, dynamic environments and interact with the rough-and-tumble physical world. “It’s amazing to be part of a project that’s pushing boundaries,” said Centro SIMES 11th Grader Diego Milano. “Working with Isaac GR00T and Baxter gives us access to the best technology available. I can hardly believe high school students like us are contributing to such cutting-edge advancements.”

Milano, who is responsible for training and fine tuning the AI model shared his excitement, saying, “We achieved incredible results…it gets a very good performance of the model. Our goal is to use this model with the Baxter robot at the Universidad Concepcion.”

The work being done by these high schoolers at Centro SIMES is truly groundbreaking in the field and places them near the forefront of robotics and artificial intelligence. In the same realm where universities and tech companies with access to advanced infrastructure and substantial funding are facing the same significant hurdles in Sim-to-Real training, creating robot digital twins, and running large-scale AI simulations, Opazo’s team of motivated youngsters is making amazing strides.

Centro SIMES aims for its own students to become recognized experts in robotic AI training, but also plans to make all of their research open source and available to any student anywhere through an online platform under development. Opazo has been greatly inspired by NVIDIA’s commitment to making autonomous cobots a reality with their Gear Lab initiative. “Jim Fan is the director and founder of NVIDIA's most advanced artificial intelligence laboratory, Gear Labs. He's also one of the authors of the Eureka and DrEureka algorithms, which he used to create NVIDIA's GR00T project,” Opazo explains “I was inspired by NVIDIA's Gear Lab and so I created our own Gear Lab SIMES.”

Amazingly, Opazo’s high schoolers are achieving such strides on an absolute shoestring. AI training, particularly for robotics, requires significant computing power, often utilizing high-performance GPUs. An AI system like ChatGPT might consume roughly 100-200 watts of energy and a single GPU per query, but training AI models—especially for robotics—requires a much larger infrastructure, including clusters of GPUs working in parallel to process vast datasets and handle real-time sensory input and output. Overcoming the computational and data processing constraints is one of the key challenges in AI-driven robotics. Meanwhile, Opazo’s young team relies on cloud computing to run their models, hamstrung by the idiosyncrasies of wifi and internet access. The cost of hardware has limited the team’s ability to to scale up the program and offer more opportunities to more bright young minds. To address this, the team aims to acquire the “TinyBox Green V2”, a budget-friendly super PC designed for artificial intelligence applications. This compact powerhouse is equipped with four NVIDIA RTX 5090 GPUs, the latest and most powerful graphics processing units for AI. Centro SIMES is seeking investment to provide the necessary computing power to run the latest in AI models, including large language models similar to DeepSeek, as well as perform physically accurate simulations and train robots in real-time, ultimately advancing the students' work in AI and robotics.

Despite these challenges, Centro SIMES students are not just participating in advanced research and competing internationally at the highest levels—they are breaking new ground in AI and robotics, showcasing ingenuity and technical skills to rival leaders in the field. These young people’s ability to surmount such hurdles without the resources available to universities and tech companies underscores these high schooler’s determination, as well as the transformative potential of youth-driven innovation.

Centro SIMES is an advanced educational institution based in Santiago, Chile, specializing in STEM education for students aged 12 to 18. Often referred to as a “mini-university for high schoolers,” it offers high-level courses in areas such as robotics, artificial intelligence, biotechnology, and aerospace engineering, often paralleling graduate level work at top research institutes. Centro SIMES has achieved remarkable milestones, including sending student-designed experiments to space and developing particle physics projects recognized by CERN. With a mission to push the boundaries of STEM education, Centro SIMES is building a platform to offer online programs accessible to any student, anywhere, of any age or background, fostering a global community of a new generation of innovators.

NVIDIA’s Isaac GR00T software, offered freely to advance AI and robotics globally, underscores NVIDIA's strategic investment to drive widespread adoption of its GPU hardware solutions, essential for running sophisticated AI models. Centro SIMES extends its gratitude to NVIDIA for their groundbreaking contribution to robotics AI, allowing students unprecedented access to industry-leading technology and preparing them as future leaders in artificial intelligence and robotics. “Our students are working on a cutting-edge Sim-to-Real project that demonstrates extraordinary initiative and technical sophistication,” said Opazo. “This experience not only places our students at the forefront of robotics education but aligns with NVIDIA’s vision of becoming the foundational 'Android' for global robotics AI.”

For further information, please contact:

Nancy Langdon

Email: RotaryYouthExchange5280@gmail.com

Phone: 310.714.9886