iFlytek-backed robot startup bets on world models, body awareness and less real-robot data
量子位 · wechat · 2026-07-23
A QbitAI long read says a newly founded robot company from iFlytek is betting that the real bottleneck in embodied intelligence is not hardware, but the robot brain. The article argues that VLA is not the endgame, and that world models still need a missing ingredient: proprioceptive self-awareness, or what the piece calls "body cognition."
It breaks the system into three parts:
- Architecture: instead of plain VLA, the new model adds a video-generation module for future-state prediction, and treats inverse kinematics as a core task to teach the model what the body can actually do.
- Brain-body coordination: understanding, prediction, and action are trained as separate modules and then jointly aligned so planning and execution happen in tighter real-time cooperation.
- Data strategy: nearly half of training data is spatial reasoning, perception, and task-planning data, not just robot trajectories, to compensate for the scarcity of real robot data.
The article also reports strong benchmark results on LIBERO-Plus and RoboTwin 2.0, and says the company already has a real-robot deployment path through its sister robotics business.
More from Embodied
- RealSense D585 Pro AI-Powered 3D Stereo Depth Camera Unboxed for Physical AI — chrismatthieu · 2026-10-26
- Physical world agent swarm demo shared on X — pduan · 2026-10-26
- Looking for a wearable mic with best battery life that feeds recordings to personal agents via MCP — altryne · 2026-10-26
- 'Lamp Swarm': a desk of lamp-shaped robots shows what physical AI work looks like — dee_hw · 2026-10-26
- Tencent's HiSTEVE 1.0: hierarchical VLA agent plays Minecraft with native keyboard and mouse, hits 85% SOTA — _akhaliq · 2026-10-26
- SJTU's VaM lets robots learn online from mistakes, 95%+ success over 4-hour autonomous runs — chris_j_paxton · 2026-10-26