Компания SynDexy поставила первую небольшую партию человекоподобных роботов C1 для сборки прецизионной электроники
Read more
Pandaily
pandaily.com

Компания SynDexy поставила первую небольшую партию человекоподобных роботов C1 для сборки прецизионной электроники

Beijing-based embodied AI company SynDexy announced on October 30 that it has successfully completed the initial small-batch shipment of its SynDexy C1 dexterous humanoid robot. This marks a transition for the product from the research and validation phase into engineering delivery and actual deployment.

These robots are intended for executing precise assembly procedures within customer factories specializing in consumer electronics and electronic components, tasks that require a high degree of consistency and accuracy. During preliminary testing, SynDexy reported that the C1 unit achieved a success rate exceeding 98 percent when performing critical operations.

The C1 model is a wheeled, two-armed humanoid. Its primary advantage lies in integrating both visual and tactile capabilities. Hardware-wise, it incorporates proprietary visuotactile sensors, which the company claims provide micron-level spatial resolution and highly dynamic fine sensing, enabling the robot to grasp objects dexterously and handle them compliantly in real time. Software functionality is managed by SynDexy M1, a manipulation model that merges data from vision, touch, and force. SynDexy refers to M1 as the first model of its kind, although this assertion has not been substantiated with external benchmarks.

The robot possesses 22 degrees of freedom throughout its structure and operates its motion control system on a 1-millisecond cycle. The development cycle was notably brief; work on C1 commenced in April 2026, the first prototype was finalized in July, and the initial small batch was dispatched in October, totaling approximately six months. The company credits this rapid progress to establishing the product requirements based on genuine production line demands from the outset.

Founded in 2025, SynDexy develops its own robotic hardware, visuotactile sensing technology, and manipulation models. The company targets high-precision manufacturing sectors, including automotive parts, semiconductor micro-components, and consumer electronics. Fang Bin, the chief scientist, holds a professorship at the School of Artificial Intelligence at Beijing University of Posts and Telecommunications, bringing over a decade of experience in embodied perception, interaction, and dexterous manipulation. The core development team comprises members from institutions such as Tsinghua University, Beihang University, Harbin Institute of Technology, and Beijing University of Posts and Telecommunications, alongside contributions from Huawei, DJI, and Tencent, with researchers constituting over 90 percent of the workforce.

Handling delicate items like small connectors and flexible materials in electronics plants presents a significant challenge for robotics due to part variability and minimal tolerance for error. Touch sensing is identified as a key method through which industries are attempting to overcome this difficulty.

SynDexy intends to maintain its focus on high-precision dexterous manipulation within industrial applications. The company chose not to disclose the identities of the customers who received the initial robots or the exact quantity delivered.

Similar stories

Midea MIRO U robot installed screws on 100,000 washing machines in three months
Read more
pandaily.com

Midea MIRO U robot installed screws on 100,000 washing machines in three months

The six-armed humanoid robot MIRO U from Midea has worked at a washing machine manufacturing plant in Uxi over the past three months, where it performed the installation of screws on the rear panel of 100,000 such machines.

These figures were presented by the head of the embodied robotics innovation center at KUKA during a media tour organized by the Guangdong Department of Industry and Information Technology in mid or late October, according to a report by Time Finance published in QQ News.

At the Midea Global Innovation Center laboratory in Shunde, Foshan, journalists observed MIRO U screwing in bolts at an average speed of six bolts in about 10 seconds, with a reported success rate of 99.8 percent.

Midea, which owns the German robot manufacturer KUKA, considers the six-armed solution practical. Two-armed humanoid robots possess flexibility, but their efficiency is only about 30 percent of human efficiency, which is insufficient for a home appliance production line. The use of six arms on one machine allows its productivity per square meter to match that of skilled workers.

Midea first introduced MIRO U in December 2025 as the third generation of its humanoid robot lineup. This wheeled robot is capable of autonomously climbing vertically and rotating 360 degrees in place. The company follows an innovative approach: robots are first tested at its own washing machine manufacturing plants in Jingzhou and Uxi, as well as at the logistics warehouse in Wuhan, before being offered to external customers.

MIRO U is part of a broader provincial trend. According to a report, by September, more than 2,000 humanoid robots had been deployed in Guangdong's manufacturing companies operating in real production environments, and it is expected that this number will exceed 3,000 by the end of the year.

Other examples from the tour demonstrate the transition of humanoid robot work from demonstrations to measurable tasks. At the headquarters of X Square Robot in Shenzhen, a two-armed robot turns parcels so that their labels face up, sorting over 1,800 packages per hour. The company AI² Robotics reports that its robots are used in semiconductor display, automotive, and biopharmaceutical enterprises, but managers still consider full-scale implementation a distant future.

The head of Huayan Robotics noted that solutions that perform well in simulation often lose accuracy in real workshops due to changes in lighting, component variations, vibration, oil pressure, and cycle time constraints.

Official Guangdong data shows that in 2025, the province produced 336,300 industrial robots, accounting for 43.5 percent of the national volume, and the Pearl River Delta is capable of creating prototypes within 48 hours and delivering small batches within seven days.

Shenzhen startup Primotion unveils Paimon-0 model, reducing action inference time to 300 ms
Read more
pandaily.com

Shenzhen startup Primotion unveils Paimon-0 model, reducing action inference time to 300 ms

Primotion, a Shenzhen-based startup founded in July 2026 to develop chips for embodied AI, has entered the public sphere not with silicon, but with a model. Its embodied foundational model, Paimon-0, ranked first in several publicly available robotics benchmarks, according to a detailed report by QbitAI published on Tencent News, as well as an earlier report from The Beijing News.

The company claims that Paimon-0 achieved a score of 0.7754, leading the RoboColiseum leaderboard, where it also led in the sub-rankings of general manipulation and disturbance adaptation. The model scored 58.1 in RoboCasa365 and demonstrated an average success rate of 0.73 on VLA-Arena. Furthermore, it won a policy competition for two-armed robots with a score of 46.599. All these results are presented by the company itself.

The model is built around a world action model that predicts future states, which are then translated into actions using a reverse dynamics model. Since rendering the full future video is too slow for real-time robotics, Paimon-0 uses query tokens to extract only essential features, such as object 3D pose, semantics, and usability, before predicting actions on the device. According to Primotion, this reduces the inference time for one action block from approximately 3000 ms to 300 ms. The company emphasizes that this refers to the block inference time, not the low-level control loop period. Pre-training relies heavily on first-person human manipulation videos provided by partner EgoScale.

CEO Ren Qingyuan, formerly Vice President of RISC-V processors, explains why the chip company started with a model. He believes the team must first complete the full cycle of data, model, agent, and evaluation on real customer tasks, and then use these workloads to define the chip. Primotion separates the work between a cloud planner, which they call the meta-brain, and an on-device local brain that processes high-frequency feedback from vision, touch, and force.

Hardware development has already begun. By adding an FPGA to the robot's main controller to offload pre- and post-processing, Primotion reports reducing the worst-case error in the 1 kHz EtherCAT control loop from 30 microseconds in software to 152 nanoseconds. Most of this logic is intended for transfer to their proprietary chip, which the company plans to start shipping in 2028.

The company states that it will not be creating robot casings. It plans to release the source code of its progressive Paimon-Eval model in November or December, and a second foundational model trained on first-person data is expected in about three months.

Access to pandaily.com website is restricted using Cloudflare
Read more
pandaily.com

Access to pandaily.com website is restricted using Cloudflare

The owner of the pandaily.com website has temporarily blocked user access to the resource. This restriction is implemented using the Cloudflare service.

For more detailed information about this situation, users can refer to the official Cloudflare documentation on HTTP status codes, specifically at the link https://developers.cloudflare.com/support/troubleshooting/http-status-codes/cloudflare-1xxx-errors/error-1015/.

Popular