Embodiment‑agnostic robot brain for humanoids, wheeled/mobile platforms, and manipulators

We’re building a single, embodiment‑agnostic robot brain that learns task and world intelligence from large‑scale human and robot data, then plugs into humanoids, wheeled/mobile platforms, and manipulators via lightweight adapters—powered by the NVIDIA stack and designed for real‑world deployment.

Generalising Manuplation across embodements

Generalising Manuplation across embodements

Generalised grasping, placing, and contact‑rich tasks across arms, tabletop robots, and humanoids.

Generalised grasping, placing, and contact‑rich tasks across arms, tabletop robots, and humanoids.

Generalising locomotion across embodements

Generalising locomotion across embodements

Generalised navigation and movement across bipeds, wheeled robots, and other mobile platforms.

Generalised navigation and movement across bipeds, wheeled robots, and other mobile platforms.

Why One Brain for All Robots?

Today, every new robot body—humanoid, wheeled, or arm—requires rebuilding intelligence from scratch. We unify this with one embodiment‑agnostic brain that generalizes across morphologies, cutting deployment time, data cost, and engineering effort while enabling true multi‑robot fleets.

Today, every new robot body—humanoid, wheeled, or arm—requires rebuilding intelligence from scratch. We unify this with one embodiment‑agnostic brain that generalizes across morphologies, cutting deployment time, data cost, and engineering effort while enabling true multi‑robot fleets.

Grounded in an
open robotics stack.

Grounded in an
open robotics stack.

Our initial experiments build on existing open models and simulators rather than claiming to train a foundation model from scratch.

NVIDIA Isaac GR00T

Open robot foundation models for generalized humanoid reasoning and skills. We are evaluating GR00T as a high-level, multimodal task-policy starting point.

NVIDIA Isaac GR00T

Open robot foundation models for generalized humanoid reasoning and skills. We are evaluating GR00T as a high-level, multimodal task-policy starting point.

Isaac Lab + Isaac Sim

Simulation and learning environments for policy training, evaluation, and systematic sim-to-real experimentation. The first prototype is designed to run here.

Isaac Lab + Isaac Sim

Simulation and learning environments for policy training, evaluation, and systematic sim-to-real experimentation. The first prototype is designed to run here.

Teleoperation and generated trajectories

Small sets of targeted demonstrations plus simulation-derived data are the planned path to adapting a shared system to a particular robot body and task.

Teleoperation and generated trajectories

Small sets of targeted demonstrations plus simulation-derived data are the planned path to adapting a shared system to a particular robot body and task.

Soul Robotix

General-purpose approach to physical intelligence designed to learn through interaction with the world

Socials

Twitter (X)

Team@soulrobotix.com

Soul Robotix

General-purpose approach to physical intelligence designed to learn through interaction with the world

Socials

Twitter (X)

Team@soulrobotix.com

Soul Robotix

General-purpose approach to physical intelligence designed to learn through interaction with the world

Socials

Twitter (X)

Team@soulrobotix.com