LEXSUS robots grill live for 550K viewers, self-correcting failures to prove cross-embodiment AI
创业邦 · wechat · 2026-09-20
On Sept 16, Chinese startup 乐享科技 (LEXSUS) livestreamed two robots running an outdoor BBQ stand, drawing 5.5M cumulative views. A chef and a waiter robot served three customer groups for nearly an hour, repeatedly failing—dropping enoki mushrooms, handling mid-task drink requests and extra orders—yet self-correcting without any teleoperation, preserving order state and re-sequencing tasks.
- Zero-shot test: follows an August no-edit housework demo; the livestream was staged as a response to accusations of cherry-picked footage.
- Cross-embodiment: the robots were competitor hardware, deliberately chosen to prove the company's 'Yita' generalist embodied model transfers across different bodies, mapping goals to each body's capability limits.
- Brainstem architecture: layered bio-inspired design spanning intent/task planning, perception, motor planning and a 'spinal' layer for force control and collision avoidance, combining active perception, multimodal understanding, streaming memory, motor dynamics and self-iteration.
- Low-cost training: only 4B parameters, trained on 200 hours of human video with zero real-robot data.
- Business thesis: transferable embodied brains may shift industry competition from building more hardware to whose model runs on more bodies.
More from Embodied
- Open-source dual-arm robot solves tangram puzzles with 90 demos, ~60% success — DanGrover · 2026-11-12
- Robotics has three regional playbooks: China volume, Europe inspection, US pilots — ingliguori · 2026-11-12
- 3,500 Laser-Armed AI Submarine Robots Patrol Norwegian Salmon Farms — philfung · 2026-11-12
- Model Autonomously Designs a Surgical-Robot-Style Robotic Arm CAD After 75 Minutes — _Stocko_ · 2026-11-12
- Dev releases local world model turning any image into interactive video, runs on RTX 3060 — lucidml_lover · 2026-11-12
- RoboRisk benchmark shows foundation models pose high harm rates in robotics; LEASH harness cuts risk — UoLiverpool · 2026-11-12