DeepSeek invests $20.8 million in Unitree's Shanghai IPO

By Eduardo Baptista

BEIJING, Aug 6 (Reuters) - Chinese artificial quality startup DeepSeek has invested 140.8 cardinal yuan ($20.8 million) successful robot shaper Unitree and ‌agreed to jointly make AI models for humanoid machines, according to ‌a stock-exchange filing.

DeepSeek received 933,399 Unitree shares, representing 2.31% of the shares allocated done the Shanghai ​initial nationalist offering's strategical placement.

The companies, some based successful China's eastbound tech hub of Hangzhou, said they would harvester DeepSeek's expertise successful AI models with Unitree's enactment successful mechanical engineering, question power and embodied intelligence.

Unitree, besides known arsenic Yushu ‌Technology, would springiness penchant to ⁠DeepSeek erstwhile procuring model-training services and method solutions. DeepSeek would likewise favour Unitree erstwhile purchasing robots oregon exploring embodied-AI applications, ⁠the filing said.

The concern targets a cardinal bottleneck for humanoid developers: gathering a robot "brain" susceptible of knowing unfamiliar surroundings and reliably turning instructions into carnal actions.

China has ​developed comparatively ​inexpensive robots susceptible of walking, moving ​and performing choreographed routines, but utile ‌autonomous enactment requires models trained connected scarce physical-world data, including demonstrations of manipulating objects and responding to changing environments.

Chinese robot makers are operating data-collection facilities successful which humans repeatedly usher humanoids done tasks specified arsenic folding apparel and opening doors.

DeepSeek's AI models person performed powerfully connected coding, mathematics and reasoning tests, ‌while competing aggressively connected price.

But the company's ​strengths person been concentrated successful language-model bid alternatively ​than a multimodal strategy capable ​to process imaginativeness alongside text. DeepSeek has released abstracted vision-language ‌research models, including VL2, but they ​are not integrated ​into its flagship commercialized model.

A DeepSeek researcher concisely published a task successful April connected improving ocular reasoning done points and bounding boxes earlier the ​paper and authoritative GitHub ‌repository were withdrawn without a nationalist explanation.

The Unitree concern could supply ​DeepSeek with robots and physical-world information to accelerate specified work.

(Reporting by ​Eduardo Baptista. Editing by Mark Potter)