NVIDIA Isaac Sim
NVIDIA Isaac Sim is an open source reference robotics simulation application built on NVIDIA Omniverse, enabling developers to design, simulate, test, and train AI-driven robots in physically realistic virtual environments. It is built atop Universal Scene Description (OpenUSD), offering full extensibility so developers can create custom simulators or seamlessly integrate Isaac Sim's capabilities into existing validation pipelines. The platform supports three essential workflows; large-scale synthetic data generation for training foundation models with photorealistic rendering and automatic ground truth labeling; software-in-the-loop testing, which connects actual robot software with simulated hardware to validate control and perception systems; and robot learning through NVIDIA’s Isaac Lab, which accelerates training of behaviors in simulation before real-world deployment. Isaac Sim delivers GPU-accelerated physics (via NVIDIA PhysX) and RTX-enabled sensor simulation.
Learn more about NVIDIA Isaac Sim
NVIDIA Isaac Lab
NVIDIA Isaac Lab is a GPU‑accelerated, open source robot learning framework built on top of Isaac Sim, designed to unify and simplify robotics research workflows such as reinforcement learning, imitation learning, and motion planning. It leverages realistic sensor and physics simulation to support accurate training of embodied agents, providing ready‑to‑use environments, spanning manipulators, quadrupeds, and humanoids—with support for 30+ benchmark tasks and integration with popular RL libraries like RL Games, Stable Baselines, RSL RL, and SKRL. Isaac Lab features a modular, configuration‑driven design that enables developers to easily create, modify, and scale learning environments; it also supports collecting demonstrations via peripherals (gamepads, keyboards) and allows custom actuator models to facilitate sim‑to‑real transfer. The framework is built for both local and cloud deployment, accommodating flexible scaling of compute resources.
Learn more about NVIDIA Isaac Lab
Gemini Robotics 2.5
Gemini Robotics 2.5 is a vision-language-action model designed to translate visual and language inputs into actions for robots. It supports robotic systems ranging from tabletop and bi-arm platforms to humanoid bodies, including manipulation with hands and grippers and coordinated whole-body movement. The model is intended for robotics developers, research teams, and organizations working with robot hardware, helping robots interpret their surroundings, follow multi-step tasks, and handle varied objects and changing environments. Google DeepMind describes applications such as sorting and kitting parts in automotive production, handling delicate or flexible objects, and performing everyday manipulation tasks. Testers can request early access through a waitlist.
Learn more about Gemini Robotics 2.5
Gemini Robotics 2
Gemini Robotics 2 is Google DeepMind’s intelligence layer for adaptable robots, bringing whole-body control, advanced dexterity, embodied reasoning, and multi-robot collaboration to physical AI. It includes three models. Gemini Robotics 2 is a vision-language-action model that converts visual and language input into motor control, enabling humanoids and bi-arm robots to act from feet to fingertips. It can coordinate walking, crouching, reaching, balancing, and object manipulation, while controlling five-fingered hands or standard grippers for delicate and precise tasks. Gemini Robotics ER 2 serves as the high-level brain, communicating with people, understanding its surroundings, planning multi-step tasks that last several minutes, coordinating actions with the VLA, tracking progress, self-correcting failures, and allowing different robots to work together.
Learn more about Gemini Robotics 2