Dongyi Robotics shows 100x efficiency gains with 4-layer embodied AI stack at Canton Fair
智东西 · wechat · 2026-10-18
Dongyi Technology (PHYBOT) demonstrated multi-scenario robot services at the 140th Canton Fair: badminton/ping-pong rallies, bilingual guided tours, heavy transport, inspection, and dexterous manipulation.
- It migrated badminton rally capabilities to ping-pong in under a month, addressing the industry's "one robot, one training run" problem.
- Its EmbodiedPhyStack splits the stack into four layers — LLM (deep reasoning), PhyAgents (real-time edge decisions via VLA), PhyCore (motion control), and the robot body. LLM-only planning took 3 hours for a grasping task; adding PhyAgents cut a 3-object收纳-style task to under 2 minutes, a 100x+ efficiency gain, and model-driven sim training boosted new-scenario development tens of times.
- Commercial traction: a mall badminton pop-up served up to 1,000+ visitors/day (10,000 total); the quadruped carries 210kg continuously (500kg peak); the 23-DoF PhyHand dexterous hand debuted alongside the standardized PhyAgentsHub deployment solution.
More from Companies & People
- Anthropic nearly stops hiring junior staff as AI agents log 6x human hours — The Decoder · 2026-10-18
- AI art tools' biggest victims won't be Adobe, argues Reddit user — it's GIMP and open source — Interesting_Ad_5676 · 2026-10-18
- Stanford AA 274A Principles of Robot Autonomy lectures by Marco Pavone are freely available — caglar_ee · 2026-10-18
- AI virtual idol Athena is shaking up the Kpop scene — basedjensen · 2026-10-18
- Oracle Health data breach exposes medical records of nearly 20 million patients — ChuckDBrooks · 2026-10-18
- X Records 13.43B Website Visits in Q3 2026, Up 6% Quarter-over-Quarter — XFreeze · 2026-10-18