XGRIDS at NVIDIA GTC 2026: Bridging Real-World Spaces with Physical AI through Real2Sim

XGRIDS at NVIDIA GTC 2026



The excitement at NVIDIA's GTC 2026 in San Jose, California, was palpable as XGRIDS presented its innovative Real2Sim technology. This advancement addresses a critical challenge in robotics: the necessity for robots to train in simulations that accurately mirror real-world conditions. With the event running from March 16 to 19, the discussions centered on how to enhance the reliability of robotic systems through better training environments.

The Challenge of Robotics Training


To ensure robots operate successfully in real-world environments, they must be trained in settings that replicate those environments as closely as possible. This aspect was a focal point during XGRIDS’ participation, especially in their engaging Startup Pitch presentation led by CEO Sunny Liao. Liao posed a fundamental question: "How do we create training environments that genuinely reflect the conditions in the physical world?" The answer lies in a data-driven approach that combines various cutting-edge technologies.

Real2Sim: A Data-Driven Approach


XGRIDS’ solution merges LiDAR with computer vision to achieve multimodal spatial perception. This combination facilitates the high-fidelity 3D reconstruction of physical environments, thus transforming them into usable models for simulation. Compared to traditional manual 3D modeling, this strategy offers several advantages:
  • - Cost Efficiency: It significantly reduces the expenses involved in building high-fidelity environments.
  • - Continuous Updates: It allows for real-time updates as physical spaces evolve over time.
  • - Realism in Simulation: It ensures that the simulation remains as close to actual implementation as possible.

Developers at GTC 2026 highlighted that this innovative approach provides a more practical route for training and validating robotic systems, marking a significant leap forward in the field.

Integrating Spatial Intelligence in Physical AI


The presentation didn’t stop with Real2Sim. Within NVIDIA's Robotics session, XGRIDS showcased its integrated AI systems that utilize spatial intelligence within quadruped robotic platforms. This technology enables robots to continuously map and comprehend their surroundings using a complete 3D spatial structure, which enhances route planning, behavioral decision-making, and task execution. In contrast to relying solely on immediate obstacle-avoidance through local sensors, these robots can make informed decisions based on a broader understanding of their environment.

Demonstrations and Collaborations


At GTC 2026, XGRIDS also collaborated with Amazon Web Services (AWS) in a significant showcase which detailed a comprehensive workflow of Real2Sim, ranging from capturing real-world data to generating accurate world models for AI training. This collaborative effort highlighted the potential of integrating Real2Sim technologies with cloud-driven capabilities, setting a new benchmark for future robotics applications.

Looking Ahead


The long-term vision of XGRIDS is unwavering: to build an infrastructure of spatial intelligence that seamlessly transforms real-world environments into models that AI systems can analyze and train on effectively. The strides made at GTC 2026 represent a critical milestone in this journey to enhance physical AI systems. As these integrated AI systems move from experimental labs to practical applications in warehouses, cities, and construction sites, the demand for scalable and precise digital representations of environments will only grow.

XGRIDS is building the essential bridge from real-world capture to simulation, enabling a new era of robotics that will revolutionize how we interact with technology in our daily lives.

For media inquiries, please contact: [email protected]
Visit XGRIDS for more information
XGRIDS Image

Topics Consumer Technology)

【About Using Articles】

You can freely use the title and article content by linking to the page where the article is posted.
※ Images cannot be used.

【About Links】

Links are free to use.