NVIDIA is assembling a coalition of robotics companies to tackle the next frontier in AI: teaching humanoid robots to grasp, manipulate objects, and use tools with human-like dexterity. The effort signals a strategic pivot in how the chip giant views the future of physical AI, moving beyond chatbots and vision systems to focus on the embodied skills that will determine whether robots can actually perform useful work in factories, kitchens, and warehouses.

What Is RLDX-1 and Why Does It Matter?

RLWRLD, a startup developing robotics foundation models, unveiled RLDX-1 at an event called "Dexterity Night in SF" on June 19, 2026. The model is purpose-built to help humanoid robots perform contact-rich tasks such as grasping, pouring, and tool use. Unlike earlier robot AI systems that focused primarily on vision and language, RLDX-1 was designed from the ground up with what the company calls a "Dexterity-First" philosophy.

The model has already demonstrated its capabilities across multiple real-world scenarios. RLWRLD reported benchmark results in humanoid tabletop manipulation, kitchen tasks, and real-world coffee-pouring evaluations. Importantly, RLDX-1 runs across multiple robot embodiments, including WIRobotics' ALLEX humanoid, Franka Research 3, and OpenArm, suggesting the model can generalize across different robot hardware.

"Robot AI so far has been stuck on 'seeing' and 'talking.' If robots are going to do real work in factories, kitchens and warehouses, they need to grasp, feel and hold on. RLDX-1 was built from day one to fill that gap," said Junghee Ryu, CEO of RLWRLD.

Junghee Ryu, CEO of RLWRLD

How Is NVIDIA Supporting the Physical AI Ecosystem?

NVIDIA's involvement goes beyond cheerleading. The company provided the entire technical foundation for RLDX-1's development. This includes NVIDIA Isaac GR00T (a robotics foundation model framework), NVIDIA Isaac Lab (a simulation environment), NVIDIA Isaac Sim (for virtual robot training), and cuRobo (a tool for collision-free motion planning). For the heavy computational lifting, RLWRLD used NVIDIA Hopper GPUs to train the model and NVIDIA Jetson AGX Thor with NVIDIA TensorRT for running inference on the robots themselves.

At the launch event, Amit Goel, Head of Robotics Ecosystem and Edge AI Product at NVIDIA, took the stage to publicly endorse the partnership. His presence and remarks underscored NVIDIA's strategic commitment to physical AI as a major growth area.

"RLWRLD is one of the core partners in the physical AI ecosystem we are building at NVIDIA," stated Amit Goel, Head of Robotics Ecosystem and Edge AI Product at NVIDIA.

Amit Goel, Head of Robotics Ecosystem and Edge AI Product at NVIDIA

What Companies Are Part of This New Alliance?

The "Dexterity Night" event brought together a coalition of hardware and software companies signaling the formation of a broader alliance around dexterous manipulation. The participants included:

• Hardware Partners: WIRobotics, Enactic, and Origami Robotics, which manufacture humanoid and robotic platforms that will run RLDX-1.
• AI Infrastructure: Proception AI and other companies providing perception and sensing capabilities for robots to understand their environment.
• NVIDIA Leadership: Senior leaders from NVIDIA attended, signaling the company's deep investment in the ecosystem's success.

This coalition structure suggests that NVIDIA is positioning itself as the infrastructure backbone for a new wave of robotics companies, much as it has done for AI research and large language models.

Why Does Dexterity Matter More Than Vision or Language?

For years, robotics research has focused on two problems: helping robots see (computer vision) and helping robots understand language (natural language processing). But as RLWRLD's CEO noted, these capabilities alone don't enable robots to do useful work. A robot that can see a coffee cup and understand the instruction "pour the coffee" still cannot execute the task without the ability to grasp the cup, feel its weight, adjust grip pressure, and pour without spilling.

Dexterous manipulation requires a different kind of AI. The robot must learn contact-rich interactions, where success depends on subtle feedback from touch sensors and proprioception (the sense of where the robot's limbs are in space). RLDX-1 addresses this gap by training on data and simulations that emphasize these tactile and force-control skills, not just visual recognition.

What Does This Mean for the Future of Robotics?

The emergence of RLDX-1 and NVIDIA's public backing suggest that the robotics industry is entering a new phase. Just as foundation models like GPT and BERT accelerated natural language AI, a foundation model for dexterous manipulation could unlock a wave of practical robot applications. Companies building humanoid robots will no longer need to train manipulation skills from scratch; they can fine-tune RLDX-1 on their specific tasks and hardware.

This shift also reflects a broader recognition within the AI industry that the next frontier is not just smarter software, but smarter embodied systems. NVIDIA's investment in the physical AI ecosystem positions the company to capture significant value as robotics moves from research labs to real-world deployment in warehouses, manufacturing plants, and service industries.