Former Nvidia AI Research VP Starts Up, Raises $90 Million to Build 'The Matrix' for Robots

Perceive Beating AI News, former NVIDIA AI Research VP Sanja Fidler, along with two long-time research partners Zan Gojcic and Huan Ling, has founded Veeda AI and secured a $90 million seed funding round, with participation from Khosla Ventures and Radical Ventures. Veeda does not focus on the robot per se, but rather on building a multi-modal world model to provide robots with a virtual training environment that can scale massively.

In reality, enabling robots to learn through trial and error is both costly and slow. A single mistake can lead to equipment damage, and training cannot be parallelized at scale like compute tasks. Veeda aims to shift trial and error into a simulated world, allowing robots to operate, fail, and retry in a virtual environment. Fidler directly refers to this as the "Matrix" of Physical AI.

Such world models not only generate realistically looking videos but also accurately simulate space, motion, and object interactions. Without this, it is challenging to reliably train robots. The three founders have extensive past experience in related research at NVIDIA, recently collaborating on projects such as Lyra 2.0 and OmniDreams involving 3D and world simulation; Fidler and Huan Ling have also been involved in the NVIDIA Cosmos World Modeling Platform.

Veeda is currently in its early stages, with specific models, clients, or product launch timelines yet to be disclosed. The official website indicates that the team is distributed across Toronto, Zurich, Mountain View, and Singapore, and is currently hiring individuals skilled in world modeling, data, and large-scale training.