Niantic, NVIDIA and Flexion Achieve Zero Real-World Fine-Tuning for Humanoid Robots
lukas_m_ziegler · x · 2026-08-11
Niantic, NVIDIA, and Flexion jointly demonstrated a breakthrough robotics training pipeline: training a humanoid robot entirely within a photorealistic digital twin, enabling it to walk in a real-world office with zero real-world fine-tuning.
Workflow Breakdown:
- Space Scanning: Niantic creates 3D scans of the physical environment using just a few minutes of 360-degree video.
- Simulation Training: NVIDIA handles the robot training inside its Isaac Sim platform.
- Sim-to-Real Transfer: Flexion bridges the gap to reality using RGB-only navigation.
This approach compresses months of on-site physical adaptation down to just days, drastically accelerating deployment for embodied AI.
Related event: Niantic and NVIDIA Achieve Zero-Shot Sim-to-Real for Humanoid Robots(2 posts)→
More from Embodied
- CoRL 2026 Announces 32 Accepted Workshops Focusing on Embodied AI Frontiers — Majumdar_Ani · 2026-08-11
- Unitree's Shanghai IPO Oversubscribed 5,526x as Physical AI Booms — 0xSammy · 2026-08-11
- Deep Dive into Humanoid Robotics Actuators: The Hybrid RV Reducer — MarwaEldiwiny · 2026-08-11
- Ant Group Leads Multi-Million Funding in Daimon Robot, Unveiling First Tactile-Grounded World Model — 量子位 · 2026-08-11
- Can a £3.5k Local AI Workstation Match ChatGPT? Reddit Debates Hardware Choices — ForeignAdagio9169 · 2026-08-11
- PA-CF Nylon Filament Proven Superior for 3D Printed Robotic Grippers — DominiqueCAPaul · 2026-08-11