Close Menu
  • AI
  • Content Creation
  • Tech
  • Robotics
AI-trends.todayAI-trends.today
  • AI
  • Content Creation
  • Tech
  • Robotics
Trending
  • Potential Cognition AI Spherical Would Elevate $2B-Plus at $75B Valuation – Unite.AI
  • OpenRouter Cuts Three Suppliers After AI Immediate Publicity Findings – Unite.AI
  • A Group of Researchers from Columbia and CertiK Introduce TensorLean for Formal Verification of GPU Kernels
  • Collectively AI Discloses Immediate Leak After Coding Agent Uncovered Credential – Unite.AI
  • Dragos Finds Immediate Leaks and Stolen Keys in AI Provide Chains
  • Lean and Mathlib Communities Certify Claude-Assisted Riemann Proof – Unite.AI
  • AWS Routes HyperPod Inference Throughout Areas From One Endpoint – Unite.AI
  • DailyAI Briefing: A Lean-checked Riemann proof, plus 4 extra
AI-trends.todayAI-trends.today
Home»Tech»Genesis AI releases Nyx, Quadrants and Genesis World 1.0 Physics Platform to Scalable Robotics Model Evaluation

Genesis AI releases Nyx, Quadrants and Genesis World 1.0 Physics Platform to Scalable Robotics Model Evaluation

Tech By Gavin Wallace30/05/20268 Mins Read
Facebook Twitter LinkedIn Email
Mistral Launches Agents API: A New Platform for Developer-Friendly AI
Mistral Launches Agents API: A New Platform for Developer-Friendly AI
Share
Facebook Twitter LinkedIn Email

Genesis AI Released Genesis World 1.0. It consists of 4 components: the Genesis World engine (physics), Nyx, a real-time rendering tool that tracks a virtual path in real time (path-tracing), Quadrants compiler (Python to GPU), and an interface for simulation. This platform was designed to help accelerate the development of robotic foundation models through simulation.

Two bottlenecks exist in the development of robotics models: Data and speed. Data has been a major focus of the field. Genesis AI argues the slower, less-discussed bottleneck is the model development cycle itself — specifically, how fast teams can evaluate candidate policies and compare model checkpoints.

What problem does it solve?

Genesis’ typical evaluation of policy involves hundreds of tasks, each with hundreds or episodes. Running that in the real world requires more than 200 hours of continuous robot operation with one operator and one robot station — for a single evaluation pass. To make statistically valid comparisons at different checkpoints you need many passes.

Genesis World 1.0 performs the same analysis in less than 0.5 hours. No human is involved and the results are bit-exact. It is approximately two orders-of magnitude faster than a real world evaluation.

It was the research team’s deliberate choice to prioritise evaluation rather than simulation in order to generate training data. They reasoned that, if the simulation distribution for training and evaluation is identical, then a performance increase could simply be a result of better fitting simulator dynamics as opposed to an actual improvement. The signal is cleaner when the pipelines are kept separate.

This evaluation method is described by the research team as zero-shot real-to-simPolicies evaluated through simulations are only trained using data from the real world. No simulated data enters pretraining.

Sim-to-Real Correlation Results

Genesis Research team has reported a Pearson correlation coefficient of 0.8996 (95% CI: [0.7439, 0.9314]The evaluation covered three model variants (Small, Medium and Large) across 14 tasks with 200 episodes per task. Three model variants were evaluated (Small. Medium. Large) for 14 tasks and 200 episodes. To estimate confidence intervals, the research team performed 1,000,000 iterations of bootstrap.

The Mean Maximum Rank Violation (MMRV) — a metric proposed in SimplerEnv — was 0.0166 (95% CI: [0.0102, 0.0474]). If the MMRV is low, the simulator will maintain the ranking relative of the performance between the different models.

Genesis developed a real time side-byside system to determine the source of divergence between sim and reality. The simulator runs in parallel with the robot, starting from the same initialization. Observations — camera frames and proprioception — can come from the simulator, the robot, or a blend of both. By switching one source, you can isolate whether the divergence is due to physics, rendering or communication.

The reality gap between the two is now a thing of the past. The 45% reduction in sizeWhen compared to the alternative simulators, FID scores on their dataset are higher.

https://www.genesis.ai/blog/the-role-of-simulation-in-scalable-robotics-genesis-world-10-and-the-path-forward?x_refresh=1

Genesis World: The Four Components

Nyx — Real-Time Path-Traced Rendering

Nyx is an GPU-accelerated route tracer which plugs in to Genesis World like a camera. The software is also available in the gs-nyx-plugin Python package. There are wheels available in x86-64 Linux, Windows 10/11 (manylinux version 2.34+), and for CUDA.

Nyx targets sound-free 1080p frame in Four minutes or less No baking, no ghosting. In order to achieve this goal, the software uses a transparency buffer, MSAA hardware ray-tracing, hardware matrixcores, video compression, and a GPU driven architecture without binding.

Construction is based on path tracing: Multi-bounce lighting and indirect illumination. The camera is then anchored to the ground. HDRI pipes light scenes using measured radiance. Internal scanning and photogrammetry provide assets. 3D Gaussian squares can be used to extend coverage when mesh reconstruction is not sufficient.

Nyx’s physics are based on batching rather than scene-byscene. This allows thousands of parallel rollouts — each with its own scenario, lighting, and camera trajectory — to pass through a single unified rendering pipeline.

Genesis World Physics — Unified Multi-Physics Engine

Genesis World physics software is open source and Apache 2.0 compatible. It runs multiple physics simultaneously: articulated rigid body (MJCF/URDF/USD), FEM, MPM, fluids (including granular and elastoplastic), SPH, fast cloth (including position-based liquids), PBD, and SPH.

Behind the same API, three interchangeable couplers can be used: A fast coupler that is general purpose; a semi-analytic primal coupler in Drake style with hydroelastic contacts; and an incremental potential contact (IPC), which allows for contact without intersection in scenes with deformability. The only change required to switch between couplers is a single-line of code. Assets, sensors or the interface for policy are not affected.

Genesis World 1.0 features two new Solvers. Genesis World 1.0 introduced two new solvers. External Articulation Restraint The second is a built-on-top-of-libuipc solution that embeds joint space dynamics into IPC’s optimization. This means joint-space and contact forces are resolved simultaneously, rather than being staggered between separate solvers. Second is barrier-free elastodynamicsIt replaces IPC’s logarithmic barriers with an augmented Lagrangian. IPC standard’s barrier leads to a Hessian that is ill-conditioned with tightening contacts. It is possible to avoid this problem by using the enhanced Lagrangian formulation, which enables every contact pair that has been detected through continuous collisions to be added immediately into the active sets. The research team is reporting up to 103× faster The IPC-based system offers better performance in scenes with high contact density, while maintaining intersection-free guarantee.

Other improvements include: New sensors (point cloud tactile, temperature grid, proximity), Implicit FEM solver with Newton and CG, as well as expanded support for assets (URDF xacro, MuJoCo universal actuators, compound/microscopic joints, equal/weld constraint, etc.).

Quadrants — Python-to-GPU Compiler

Quadrants was developed by Genesis AI, and is available under Apache 2.0. Kernels written in Python are JIT-compiled via LLVM to NVIDIA CUDA (or AMD ROCm), Apple Metal (or Vulkan) and x86/ARM64 CPUs. In June 2025, it was forked off Taichi. Named after the Chinese saying, “Taichi gives birth to Four Forms” (Quadrants).

Performance improvements compared to upstream Taichi 4.6× Warm-cache startup time for Genesis locomotion and manipulation benchmarks. Warm cache startup time single_franka_envs.py dropped from 7.2 seconds to 0.3 seconds — a more than 10× speedup. Autodiff in reverse-mode is now standard on all backends. This makes differentiable simulator portable.

The physics steps are stored as graphs of a single kernel, which eliminates the launch delay per step. Independent kernels run in parallel via streams. Dense linear algebra (Cholesky factorization, triangular solves) compiles to 16×16 tile-blocked code paths. The perf-dispatch is a layer that benchmarks and caches fastest choices per signature. Tensors and PyTorch share memory via DLPack, with interoperability that is zero-copy.

Installation:

Simulation Interface

Fourth, the simulation interface is a tool that integrates the physics engines, renderers, and compilers into an application-ready system. It includes photogrammetry for the creation of digital twins and an automated pipeline to generate programmatic environments.

Evaluation Framework

Genesis structure evaluation is based on a taxonomy involving orthogonal perturbation lines in approximately ten dimensions. This framework was adapted from “A Taxonomy for Evaluating Generalist Robot Manipulation Policies“:

  • Visual: Background variation, lighting, and camera distortion
  • Behavioral: Unseen objects combinations, placement variants, robot configuration
  • Semantic: language rephrasing, subtask ordering, camera viewpoint

Each axis has one variable parameter, while the rest remain nominal. The team of researchers defines robustness for a particular axis in terms of the performance that is retained when perturbed compared with the nominal setting. The number of data points required to perform perturbation sweeps over multiple training checkpoints and model scales is thousands. The only way this scale can be achieved is because the simulation evaluation works without hardware.

What you need to know

  • Genesis World 1.0 is made up of four main components: Nyx, Genesis World and Quadrants.
  • A Pearson correlation coefficient of 0.8996 was achieved with the hardware implementations for each task and episode, spanning 14 tasks.
  • No simulated data is used in pre-training; all policies are trained using real-world data.
  • Barrier-free elastodynamics achieves up to 103× speedup over traditional IPC in contact-heavy scenes.
  • Quadrants achieves up to 4.6× faster runtime versus upstream Taichi, with warm-cache startup dropping from 7.2s to 0.3s.

GitHub Repositories:

Also, feel free to follow us on Twitter Join our Facebook group! 150k+ ML SubReddit Subscribe Now our Newsletter. Wait! Are you using Telegram? now you can join us on telegram as well.

Want to promote your GitHub repo, Hugging Face page, Product release or Webinar?? Connect with us


Michal is a professional in data science with a Masters of Science degree from the University of Padova. Michal Sutter excels in transforming large datasets to actionable insight. He has a strong foundation in statistics, machine learning and data engineering.

AI physics Robotics x
Share. Facebook Twitter LinkedIn Email
Avatar
Gavin Wallace

Related Posts

A Group of Researchers from Columbia and CertiK Introduce TensorLean for Formal Verification of GPU Kernels

22/11/2026

OpenAI Ships GPT-Stay-1.1 for Voice Brokers, Cuts Idle Value to $0.02/Min

21/11/2026

Meet Pondering Machines Inkling 2 Mini: 26B Looped MoE Runs on a 32GB Mac

21/11/2026

Baseten Releases Bison-1 and Groundhog-1, Topping Choice Mannequin Index

21/11/2026
Top News

Thinking Machines Cofounder’s office relationship preceded termination

OpenAI lost 4 key researchers to Meta

Elon Musk’s Grok ‘Undressing’ Problem Isn’t Fixed

Intel takes major step in its plan to purchase chip startup SambaNova

Stealth Startup Believes it Has Solved the Memory Crisis

Load More
AI-Trends.Today

Your daily source of AI news and trends. Stay up to date with everything AI and automation!

X (Twitter) Instagram
Top Insights

AI Usually Cites Internet Pages It Has By no means Even Learn – Unite.AI

18/10/2026

Compare the 16 best generative AI coding tools in 2026: features, and which one is right for you?

24/06/2026
Latest News

Potential Cognition AI Spherical Would Elevate $2B-Plus at $75B Valuation – Unite.AI

22/11/2026

OpenRouter Cuts Three Suppliers After AI Immediate Publicity Findings – Unite.AI

22/11/2026
X (Twitter) Instagram
  • Privacy Policy
  • Contact Us
  • Terms and Conditions
© 2026 AI-Trends.Today

Type above and press Enter to search. Press Esc to cancel.