AUREVAS
01 / 11

AUREVAS

Your robot is only as good as its training.

Creating the first plug-and-play platform to train your robots as simple as possible.

NVIDIA Omniverse simulator interface with node graph and property panels

No . . .

This is not sci-fi.

This complex interface is what robotics engineers have to deal with every day.

Simulation training is going to be the standard in robotics.
But simulators like NVIDIA Omniverse are ridiculously hard to set up and use.

But why should it be?

Researchers, engineers and developers have to build and work with digital environments that are already hard to start with for any project.

Then they have to train the robots with reinforcement learning (RL) that is fragile and difficult to tune right.

That is why R&D and robotics training still take months, high costs and lots of compute.

Something needs to be done to solve this.

Robotics lab with many robot arms and mobile robots
Warehouse filled with humanoids, arms and mobile robots
Mobile robots and arms in a warehouse
Robot arm in a simulator scene
Simulated robot arm sorting blocks

Our solution

An AI copilot that sits on top of existing simulators that automates and monitors the entire process from start to end.

Aurevas AI Copilot panel running inside Isaac Sim

illustrationThe Aurevas copilot sits on top of the complex Isaac Sim. You only interact with the copilot. It’s that simple.

Making robotics training intuitive.

simple and easy as it should be in this era.

Aurevas copilot will take you through everything.

From setting up digital twins and environment to helping you orchestrate the training specifics until sim to real deployment.

So you can focus on the cool stuff when building your robots.

Aurevas copilot: new training project, proposed plan with build simulation, train candidates and compare behaviour

illustration

Reinforcement Learning is the foundation for advanced robotics.

Disney Research bipedal robotsDisney Research
Olaf robot in Disney's Kamino simulatorDisney Research
Disney bipedal robots walking with peopleDisney Research

But reward tuning and reward function design remain as the biggest bottlenecks in development and scaling.

So we spent a long time to research and test how we can solve this major issue...

And we successfully created a method to do so!

click to view our published paper in IEEE↗
83.3%

Our copilot increases the robot's learning adaptability rate by 83.3%

48.7%

We achieve 48.7% higher training performance than standard commercial robotics pipelines out of the box.

83%

We replaced blind, random AI trial-and-error with an 83% predictability rate in task execution.

Simulation of robot arms at tables
Robot arm training in simulator
Many humanoids training in parallel on a grid
Aurevas compare behaviours screen
Aurevas preference recorded screen

Sim to real

Sim to real.

Training in simulation
Real-life testing

Recognition

Published, awarded, and already in conversation with the people who build robots.

IEEE Xplore (peer reviewed paper)Published · IEEE ↗
Singapore Scientific Conference 2025Published ↗
Four international conferencesPresented
Singapore Science & Engineering Fair 2026Gold Award
A*STAR ARTC & SIMTechInnovative Coding Award
NTU Nanyang Research ProgrammeGold Award
Doosan RoboticsIn collaboration talks
Physical AI Expo, HangzhouInvited · Mar 2027
The founders at SSEF 2026 with their poster and robotic-arm prototype
SSEF 2026 — presenting the published method alongside the first robotic-arm prototype that we tested on.

And we have plenty more ideas that we are currently researching and testing.

+ future works

Team

Founders

Jovan Aurelio Hartono

Jovan Aurelio Hartono

CO-FOUNDER

Robotics Lead
and Business Development.

Google Scholar ↗ LinkedIn ↗
Parith Avasadanond

Parith Avasadanond

CO-FOUNDER

Research Lead
and Implementation Machine Learning Engineer

LinkedIn ↗

A science-and-research based approach, simplified and commercialized so that everyone can use it.

Direction.

We are looking for the four things that turn a published method into a platform people can use.

01

Mentorship & space

Technical advisors in robotics and reinforcement learning, and a physical place to build.

02

Pilot partners

Labs and teams willing to run real tasks through Aurevas and tell us exactly where it breaks.

03

Hardware access

Arms, humanoids and compute, to close the sim-to-real loop on real robots rather than in theory.

04

People

Engineers who want to build the tooling layer for physical AI alongside us.

AI robotics is inevitable. We are building the platform that lets everyone have a piece of it.

Navigate with ←→ or scroll