The Company’s High-Fidelity Multi-Body Dynamics Solver Targets the Data Quality Bottleneck at the Core of Robot Learning – and Lays the Groundwork for Physics-Native World Models
MONTRÉAL, QC / ACCESS Newswire / May 13, 2026 / Uncharted Dynamics today announced the closing of a multimillion-dollar seed round led by K2VC. The capital will accelerate development of the company’s proprietary multi-body dynamics solver and its expansion across the North American robotics market.
The Bottleneck Isn’t Compute. It’s Physical Consistency.
The scaling hypothesis that powered large language models doesn’t transfer cleanly to embodied AI. More data, larger models, and more compute can amplify capability – but only if the underlying data is physically consistent. In robotics, it often isn’t.
What determines task success is rarely what the camera sees. It’s what happens at contact: how force propagates through a grasp, how friction evolves across a surface, how a deformable object responds to load. These signals are either absent, simplified or systematically approximated in most simulation pipelines – and the errors compound downstream.
“We’re not arguing against scale,” said Zhewen He, CEO of Uncharted Dynamics. “We’re arguing that scale requires a physically accurate foundation first. Otherwise, more data just amplifies a systematic error.”
A Physics Solver Built for Contact-Rich Robotics:
At the core of Uncharted Dynamics‘ platform is a high-fidelity multi-body dynamics solver engineered for the failure modes of mainstream simulation tools: deformation, soft-rigid coupling, soft contact, and contact-rich manipulation. Most off-the-shelf engines handle rigid-body kinematics reasonably well. They lose fidelity exactly when it matters most – dexterous manipulation, soft object handling, long-horizon assembly tasks.
A physics solver is not about rendering. Where a graphics engine computes what something looks like, a physics solver computes why a specific outcome occurred: how force moved through a system, how contact evolved, how material state changed. For a robot, that causal chain – not the pixel sequence – is what needs to be learned.
The company calls the output Physics-Augmented Data: datasets enriched with causal physical labels that cameras cannot observe and sensors cannot collect at scale. These include complete 6-DoF contact wrenches, deformation feedback, friction characteristics, and material response – the signals that explain not just that an object moved, but why.
This is also a defensible position at the infrastructure layer. Physics solvers of this fidelity require years of domain-specific R&D to build and validate. The company that owns physics ground truth at scale owns a compounding data asset – one that becomes more valuable as robot learning shifts from imitation toward generalizable prediction and planning.
From Infrastructure to World Models:
The near-term focus is in data and simulation. But the strategic horizon is larger.
As robotics moves toward world models – systems that allow robots not just to imitate actions, but to predict, plan, and reason about physical outcomes – the quality of the model’s internal physics becomes the binding constraint. A world model with approximate contact dynamics will hallucinate physical states. More compute makes it hallucinate faster.
Uncharted Dynamics is positioned to be the physics substrate for that next generation: not just a data vendor, but the layer that makes physically grounded world models possible. The solver, the datasets, and the evaluation infrastructure the company is building today are the foundation for a physics-native model stack – one where simulation fidelity and model capability compound together.
“Some companies are building bodies. Some are training brains,” He said. “We are building the physically reliable classrooms those brains have to grow in.”
Team:
Based in Montreal – home to one of the world’s densest concentrations of AI research talent through Mila, McGill, and Université de Montréal – Uncharted Dynamics has assembled a team at an uncommon intersection. CEO Zhewen He brings a background in computational neuroscience and direct experience in the scaling phase of large language models. The core R&D team specializes in multi-body dynamics, robotics simulation, and reinforcement learning, with prior project work spanning Bombardier, Toyota, and the Canadian Space Agency.
The combination is rare: researchers who understand how data shapes model capability, working alongside engineers who can model the physical world from first principles.
“From language models to embodied intelligence, the throughline is the same: a system’s capability is bounded by the structure and authenticity of the signals it learns from,” He said. “For Physical AI, those signals are force, friction, deformation – the causal structure of the physical world. Get those wrong in simulation, and no amount of data or architecture could possibly fix it.”
About Uncharted Dynamics:
Uncharted Dynamics is a foundational infrastructure company for Embodied AI, based in Montreal. The company develops high-precision multi-body dynamics solvers and physics-grounded data systems that enable physically valid robot learning. By grounding simulation in industrial-grade physical modeling, Uncharted Dynamics aims to provide the physics truth layer – and ultimately the world model substrate – for the next generation of general-purpose robots.
Media Contact: [email protected]
SOURCE: Uncharted Dynamics
