Ropedia Launches HOMIE Gen2, Betting Robots Need Scaled Human Experience, Not Bigger Models
机器之心 · wechat · 2026-08-31
Singapore-based Physical AI company Ropedia released HOMIE Gen2, its fourth-generation multimodal capture device in twelve months, built around what it calls the Human Experience Scaling Law: robot capability scales with high-quality human experience, not larger models.
HOMIE Gen2 features 4-camera 360° vision and 4-channel spatial audio, weighs 380g, runs 13 hours, with 50µs sensor sync. Specs: 0.2% relative spatial error, <2.5% depth error, 4.68mm hand-tracking error, 96% action semantic labeling accuracy—training-ready out of the box. Deployment is 10x faster than the prior generation at 1/12.5 the cost. Its data stack covers 10+ modalities (stereo video, depth, body keypoints, 3D Gaussians, 4D tracking) unified on one timeline and coordinate system. The earlier flagship dataset Xperience-10M holds 10M real-world interactions and 10k hours of egocentric video, nearly 1PB total.
CEO Chen Zhaoxi positions Ropedia as infrastructure rather than a data vendor, aiming to become a robot-learning infrastructure company in 3–5 years; co-founder and NTU professor Liu Ziwei argues bodyless capture frees data capacity from robot fleet size to human population: "Don't just build bigger brains—build better experience loops."
More from Embodied
- Humanoid robot makes ice cream at Shanghai DQ — CyberRobooo · 2026-08-31
- US builds drone barriers, but China's scale may bypass them — TechCrunch AI · 2026-08-31
- VLAct: Representation-Centric Continued Pre-training for Vision-Language-Action Models — StarVLA · 2026-08-31
- PonderPounce: Pretrained MLLM as Episode Context Engine for Robot Control — maum-ai · 2026-08-31
- Team builds robotics platform from scratch using Rust — MatthewChang · 2026-08-31
- Clone Robotics builds 'Synthetic Human Androids' using fluid-driven Myofibers instead of motors — CyberRobooo · 2026-08-31