Tsinghua wires a fruit fly connectome into robots, boosting maze success by 30 points
量子位 · wechat · 2026-11-04
A Tsinghua team built the adult fruit fly FlyWire connectome (138,584 neurons, 3.7M directed connections) into a fixed recurrent network core for embodied control. Its advantage emerges under change: matching an MLP on flat steady walking, but winning on obstacle crossing, smoother turning (29.9% lower command roughness), and speed tracking. The same connectome transfers across biped to hexapod bodies and tops seven baselines in mazes — lifting complex-route success from 20.4% to 50.6%. Mechanistic analysis shows mechanosensory and ascending pathways are spontaneously recruited for control, suggesting evolved wiring can serve as a reusable motor-control prior. Code is open-sourced.
More from Embodied
- Baidu's Apollo Go wraps two-week public robotaxi test rides in Tung Chung, Hong Kong — Baidu_Inc · 2026-11-04
- imec neural probe integrates photonics directly into 130nm CMOS backend — jwt0625 · 2026-11-04
- Chris Paxton: interactive robot sim is the future but glaringly data-limited — Scobleizer · 2026-11-04
- DreamX: video diffusion models as universal latent action learners for robots — kaist-ai · 2026-11-04
- RoboTryOn turns human videos into robot demos, lifting success rate from 2% to 60% — kaist-ai · 2026-11-04
- INMO Future glasses: dual 1080p displays, 125g, $799 early-bird on Kickstarter Dec 10 — Scobleizer · 2026-11-04