Qwen3-32B Has a Body Schema: Wheels Map to Heels, Grippers to Palms
Sauers_ · x · 2026-11-13
A probe experiment reveals Qwen3-32B's internal states contain a cross-body "body schema": when imagining a feather brushing robot parts, the model's internal representation aligns with matching human body locations — left wheel to left heel, camera to ear, chassis to chest, gripper to palm. Shuffled parts show much lower similarity, ruling out chance.
Asked to answer "where do you feel something" as a human, dog, or robot: adding dog front paws to a human body makes 24/24 answers say "hand", hind paws make 20/24 say "foot"; adding a human hand to a robot body makes 46% say "gripper" (vs 6% random direction), and a human foot makes 98% say "wheel" (vs 22% random). The model organizes body parts by functional role in a shared representational space.
More from Research
- Burkov's Hundred-Page Language Models Book: hands-on PyTorch LLM guide — burkov · 2026-11-13
- Study: AI startups raise 151% more in North America but no premium in Africa — danielrock · 2026-11-13
- MIT's AI copilot boosts structural buckling resistance 39% with only 15 seconds added — Dr_Singularity · 2026-11-13
- Robopedia launches Xperience 2, a 'gold standard' Physical AI dataset — liuziwei7 · 2026-11-13
- Zhejiang U & Baidu's Agent-R²: 4B agent beats larger open-source baselines on GDPval — ZhejiangUniversity · 2026-11-13
- BirdDreams driving world model generates perspective + BEV video from a single image — lealtaixe · 2026-11-13