Teaching robots to work in sand, ore, grain, and soil — the hardest materials on Earth to simulate — at interactive speed. GPU-native granular physics, built for robot-learning workflows from day one.
Mining, construction, and agriculture are racing to automate excavators, loaders, and harvesters — and every one of those machines interacts with granular material: ore, sand, grain, soil. Millions of colliding particles with emergent behaviour. The sim-to-real gap is widest exactly there — which is exactly where autonomy is worth the most.
Engineering-grade DEM captures granular physics faithfully — but at classical timesteps, closed-loop policy training is computationally out of reach.
Game-engine particle physics runs in real time, but it is not material-true — policies trained on it do not transfer to real ground.
Material-true granular behaviour AND thousands of parallel, faster-than-real-time rollouts — both at once. That is the gap we fill for robotics teams doing granular simulation.
GPU-accelerated granular physics, built for robot-learning workflows from day one — not a solver retrofitted for reinforcement learning after the fact.
Granular behaviour that stays true to the material — from free-flowing grain to sticky, cohesive ore — at interactive speed, not overnight.
Headless batch simulation, architected for large-scale parallel reinforcement learning — designed to scale to thousands of granular environments at once.
Model the actual terrain, muck pile, or bucket geometry — the real operation the robot will have to work in.
Two decades calibrating granular materials against measured site data — the kind of experience most robotics teams cannot access. We are looking to build that into a validated-material database together with partners and clients.
NDEM is built on NVIDIA Warp and coupled with Newton — we have already demonstrated two-way coupling with Newton’s Featherstone solver (a robot digging in granular media), the native path into Isaac Lab and Isaac Sim.
GPU-native kernels — the framework NDEM is written on.
Two-way coupled with Newton's Featherstone solver — robot + granular media.
The native path into NVIDIA’s robot-learning ecosystem.
VeloxSim Tech is a member of NVIDIA Inception, NVIDIA’s program for startups building on its platform. It backs the work we are already doing on the NVIDIA stack — maturing NDEM on Warp and upstreaming it toward Newton and Isaac Sim as a first-class granular solver.
The granular (DEM) and fluid (PBF) solvers behind many of the demos are open-source — released so the wider simulation and robotics community can build on them. NDEM, our real-time sim-to-real engine, builds on this same foundation.
Open-source GPU discrete-element solver for granular flow — the DEM foundation behind the hopper and drum demos.
View on GitHubOpen-source position-based-fluids solver for fluids and coupled multiphase — the physics behind the dam-break demo.
View on GitHubA beachhead in mining automation, expanding into construction and agriculture — anywhere a machine has to move a pile of something.
Autonomous excavation and loading. Perth sits at the global mining epicentre, where the largest autonomous fleets already run.
Autonomous earthmoving and site prep. Excavator-autonomy teams need exactly the digging physics we build.
Harvesting and grain-handling robotics — free-flowing granular media end to end.
A shipping track record, a blue-chip industrial base, and two decades of granular-materials engineering — built lean.
Led by Dr. Sam Wong — a PhD simulation engineer with 20+ years in DEM/CFD, bulk-materials handling, and software. Meet the team →
If you're training robots to dig, load, or move bulk material, NDEM brings real-time, material-true granular physics to your sim-to-real pipeline. Let's talk.
