WorldLabs launches R2S2R to train robots in simulation and transfer them to real hardware
量子位 · wechat · 2026-07-29
WorldLabs, after acquiring robotics startup SceniX, released Real-to-Sim-to-Real (R2S2R), a new engine for robot training and evaluation.
- Real-to-Sim reconstructs real robots, sensors, objects, and interactions into a task-aligned virtual world.
- Sim-to-Real uses that world to train and evaluate policies, then deploys them back to hardware.
- Fei-Fei Li said policies trained this way can run autonomously for one hour without real-world data.
- The system is meant to scale evaluation by reproducing success/failure conditions in simulation and screening checkpoints before expensive robot tests.
- WorldLabs says the same setup has transferred to platforms including ALOHA, RB-Y1, YAM, Flexiv, and xArm, with tasks such as box packing, cable routing, tube transfer, and cluttered single-object grasping.
- The deal also positions WorldLabs as a robotics infrastructure player: SceniX brings robotics, simulation, learning, and hardware expertise, while WorldLabs contributes generative world models and 3D reconstruction.
More from Embodied
- Origin Autonomy emerges with $20M to automate interior construction, robots running 3x human speed — lukas_m_ziegler · 2026-10-17
- One-piece 4-meter PETG bar counter shows how far large-format 3D printing has come — lukas_m_ziegler · 2026-10-17
- PayBox open-source SDK gives robots a wallet — a hexapod paid real USDC on-chain — R4nsack · 2026-10-17
- JR West to deploy giant humanoid robots this fall, cutting high-altitude rail crews by 30% — CyberRobooo · 2026-10-17
- Hobbyist models humanoid robot, hits battery space and servo torque limits — _Stocko_ · 2026-10-17
- Android robots learn to switch emotions in real time, crossing the uncanny valley — TinfoilTricorn · 2026-10-17