Dongfeng Motor will send its Xiaodong humanoid robots into factory sorting and quality-inspection roles in October 2026, begin small-batch trial production by year-end, and aims to reach human-level on-job capability by the end of 2027, according to remarks by Zhang Zhenlin, chief engineer of intelligent technology, reported by Yicai and CnEVPost during the company’s tenth science and technology innovation week.
Key points
- Xiaodong humanoid robots enter factory sorting and quality inspection in October 2026
- Small-batch trial production scheduled to begin by end of 2026
- Human-level operations targeted for end of 2027
- Yuanzai robot dog to reach dealerships for customer guidance and patrol before humanoid rollout
- Zhang Zhenlin says automotive large-model reuse accelerates embodied-intelligence commercialisation
Xiaodong moves from conference stage to shop floor
The Xiaodong platform debuted at the 2026 World Robot Conference and is now being validated on material-sorting scenarios inside automotive manufacturing. Dongfeng says the robot is progressing beyond single repetitive motions toward fully autonomous flexible operations on the shop floor. Sorting and quality inspection are the first named tasks: the robot must recognise parts, decide where they belong, and place them without a fixed script for every variant, a step up from the bolt-tightening or welding cells that have long been automated.
Yuanzai robot dog reaches dealerships first
Before the humanoid line starts, Dongfeng’s in-house Yuanzai robot dog will enter commercial use at dealerships. Yuanzai supports autonomous navigation, multimodal perception and intelligent interaction for customer guidance, campus patrol and smart-service roles, with upgrades planned toward more autonomous operation. A service adviser at a dealership in Wuhan could soon hand a visitor over to a four-legged robot that knows the showroom layout, answers model questions and escorts the customer to the right bay — tasks that today absorb staff hours but require no heavy manipulation.
Automotive large models as shared substrate
Zhang argued that automakers’ embodied-intelligence programmes let technology reuse across automotive large models, speeding commercialisation, and that future embodied products will not be limited to humanoids — vehicles themselves can be embodied platforms. The logic mirrors a pattern seen elsewhere: Japan recently launched a cross-manufacturer data initiative for physical AI development that pools sensor and control data across carmakers to train shared foundation models for robotics. Dongfeng’s approach keeps the data and the deployment inside one group, using the same perception and planning stacks that power its advanced driver-assistance systems to run both Yuanzai and Xiaodong.
For a plant manager, the October start means the first shift where a humanoid robot picks up a wiring harness, checks its connectors against a visual spec, and places it in the correct bin — or flags a defect — without a safety cage and without a pre-programmed trajectory for that exact harness variant. If the small-batch trial delivers consistent cycle times and defect-detection rates, the 2027 target would put Dongfeng among the first volume manufacturers to run humanoid robots on production tasks at human-equivalent speed and reliability.
Dongfeng has not disclosed independent benchmark results for Xiaodong’s sorting accuracy, cycle time or failure-recovery rate, nor has it published the validation protocol that will define “human-level operations” for the 2027 milestone. The company says the Yuanzai dealership rollout will precede the humanoid factory entry, but a precise date for that commercial launch has not been given.
