Tsinghua-spinoff Qianjue unveils predictive world model Polibrain, ahead of LeCun's similar work
量子位 · wechat · 2026-10-21
At the 2026 Physics AI & World Model conference in Nanjing, Tsinghua-spinoff Qianjue launched Polibrain, a predictive world model that forecasts task-relevant state changes instead of rendering full future frames, enabling a robot dog to autonomously locate and grasp trash.
- Technical pillars: polynomial representations for transferable key variables, an "Effective Degree" metric to quantify network function complexity, counterfactual-based causal memory, and a loosely-coupled architecture separating world understanding from body-specific execution across quadruped/humanoid/wheeled/dual-arm robots.
- Qianjue and Tsinghua published ICML work on latent-variable identifiability in 2025, roughly a year before LeCun team's similar "When Does LeJEPA Learn a World Model?" paper; the two learning criteria are claimed to be largely equivalent.
- Qianjue aims to be a brain provider, not a robot maker, betting that shared high-level representations unlock a cross-embodiment data flywheel—though cross-vendor data silos and real-world reliability remain unproven.
More from Embodied
- MIT CSAIL's X-Hinges 3D-Prints Everyday Objects With Built-In Sensing Across Three Motion Axes — MIT News AI · 2026-10-23
- Formic raises $50M Series B to scale physical AI across 100+ US factories — garrytan · 2026-10-23
- fal starts fine-tuning a physical robot to draw, results next week — OdinLovis · 2026-10-23
- PEARL demo: an LLM agent that checks experiment results, revises plans, and re-tests — KevinKaichuang · 2026-10-23
- CU Boulder's ATLAS Institute Opens Tenure-Track Robotics Professorship — ryosuzk · 2026-10-23
- MiTaS multi-resolution tactile imitation learning hits 80% success vs 35.5% vision-only baseline — Jan_R_Peters · 2026-10-23