Embodied intelligence enters factories and warehouses: now that the brain is smarter, what else must robots compete on?

Download the app

+1 Observation | Embodied intelligence enters factories and warehouses: Now that the brains are smarter, what else do robots need to compete on? The Paper reporter Qin Sheng 2026-11-04 14:11 Source: The Paper ∙ 10% Company > Listen to full text Font size

As questions continue to be raised about the efficiency, cost and commercialization prospects of embodied intelligence, robots in factories and warehouses are already handling some tasks in transport, storage and retrieval, and sorting.

On November 3, CeMAT Asia (the International Trade Fair for Logistics Technology and Transportation Systems) opened in Shanghai. From climbing racks and retrieving boxes in four directions to obstacle avoidance in narrow aisles and multi-robot coordination, the technological improvements exhibited by participating companies are refined down to the time consumed by a single turn and the space occupied by a single storage slot.

In May this year, Figure ran a livestream of three robots sorting parcels continuously for 200 hours; at various industry conferences, many embodied intelligence companies demonstrated tasks such as parcel grasping, flipping and sorting; Robot Era partnered with China Post and SF Express to deploy related robots at logistics centers... Warehousing and logistics have become an important application scenario for robots.

On October 19, the Ministry of Commerce and eight other departments published opinions on accelerating the advancement of science and technology in commerce, proposing to promote the implementation of embodied intelligence technology in the commercial sector and expand application scenarios such as home services and commercial logistics.

However, warehousing and logistics already have a variety of automated equipment in use, and newly entering robots must not only adapt to the site but also compare efficiency and cost with existing equipment.

The efficiency contest is detailed down to a single turn

"One is time, one is space, and another is quality." Wang Zihao, general manager of the Cainiao Logistics Technology Robot R&D Center, told The Paper that e-commerce warehouses have high order volumes, and some warehouses need to operate 24 hours a day; if a certain product fails to leave the warehouse in time, it may also affect subsequent order consolidation, transportation and other links. Equipment design must consider efficiency while minimizing the impact of a single point of failure on the entire warehouse.

The climbing robot exhibited by Cainiao this time can retrieve and place bins in four directions and supports double-deep storage, that is, placing inner and outer layers of bins along the depth of the rack. Wang Zihao explained that before retrieving the inner-layer bin, the outer-layer bin must first be moved aside; four-direction retrieval gives the robot more temporary empty slots to choose from and improves space utilization.

When traveling on the ground, a traditional right-angle turn requires deceleration, rotation and then acceleration. Wang Zihao said that by switching to curved travel, each turn can save about two seconds; as a robot turns many times within a warehouse and multiple devices operate continuously, the saved time keeps accumulating.

According to the introduction, after this upgrade, the climbing robot's operating efficiency increased by 15% and storage density increased by 20%.

Piece picking, by contrast, faces product differences. In the same order, there may be both standard products and irregular, hard-to-grasp products, placing different demands on the equipment's handling capability and cost.

This time, Yuanli Lingji exhibited its first embodied product achievement launched after merging with ATOMIX: a robot multi-agent layered operation system. Under the design of the plan, about 60% of standard, easy-to-pick products are handled by economical robotic arms, about 20% of complex products are handled by embodied robots, and the remaining about 20% of long-tail tasks are covered by humans, who also handle exceptions.

In pallet scenarios, ATOMIX uses four-way shuttles for dense storage on upper levels, while autonomous mobile robots handle first-level storage and picking and transport. According to the introduction, this "storage above, picking below" solution has been applied in many enterprises, with more than 100 logistics robots in a single project.

Yuanli Lingji booth

How do customers do the math?

"Each customer's needs for each robot are different." Ding Xia, vice president of Seer Intelligent, told The Paper that if the aisle is narrow, the robot needs to be made smaller; if the goods are large and there is insufficient turning space, the equipment may need to have lateral movement capability.

At the exhibition site, two Seer Intelligent AI delivery robots with different payloads operated mixed in the same area, demonstrating dynamic obstacle avoidance, meeting in narrow aisles and S-curve passage; the omnidirectional ground forklift autonomously identified and picked up pallets of different types, orientations and positions.

According to the introduction, Seer Intelligent's "robot brain" has been adapted to more than 400 types of core components, supporting more than 2,000 robot SKUs. Ding Xia explained that this does not mean more than 2,000 completely different robots; devices with similar appearances may correspond to different SKUs after replacing internal sensors.

Customers' requirements are "more, faster, better and cheaper." Ding Xia said functions must be able to meet complex environments, efficiency must be high, operation must be stable, and the price must be appropriate; in some industrial applications, the equipment investment payback period has already been brought down to 1 to 1.5 years.

However, robots also have current capability boundaries. In non-standard abnormal scenarios such as damaged pallets and stretch-film wrapping, robots may still need human intervention as a fallback, and the experience of skilled forklift drivers still has value. Ding Xia gave an example: a factory can hand some routine transport to robots while retaining forklift drivers to handle complex tasks—letting machines undertake repetitive, stable transport and letting people handle exceptions and experience-based judgment.

Multiple embodied intelligence products exhibited by Seer Intelligent

A relevant person in charge at Yuanli Lingji also told The Paper that price competition in the domestic logistics robot market is fierce, and profits in some projects are under pressure, but some equipment problems only emerge after running for a period of time, and large customers that value quality do not look only at price.

Different customers' needs also increase delivery difficulty. Ding Xia believes that the cost of changing customer scenarios may be too high; enterprises need to improve the speed of meeting different needs by adapting components and preparing functions in advance, so that customers can combine equipment like building blocks. "Standardizing the customization process is also a kind of standardization."

Cainiao, meanwhile, adjusts the body structure on the basis of standard vehicle models according to requirements such as customers' bin dimensions. Yang Bin, general manager of the logistics robot business unit at Cainiao Logistics Technology, said standardized products make it easier to quote, deliver and integrate, and are also more conducive to cost control.

Yang Bin stressed that the climbing robot business has just completed the stage from product launch to commercialization validation, and when choosing channel partners, the company also does not want to win orders through price competition or excessive promises on delivery time.

If the brain is smarter, can the hardware keep up?

A person in charge at Origin Intelligence mentioned earlier that in the past the industry focused more on whether software capabilities could keep up; as models improve, hardware limitations have also become more obvious.

A robot's local computing resources also have to handle tasks such as control and camera data processing, and when deploying models it is necessary to consider computing power, bandwidth and heat generation; more powerful chips may in turn significantly increase the cost of the whole machine. He said the company is exploring more cost-effective chip solutions, and model inference costs will also affect actual deployment.

Cainiao's upgrade of the climbing robot this time likewise involves both software and hardware; in addition to scheduling and path optimization, it also includes adjustments to the bin-picking structure, equipment weight reduction and battery performance improvements. Wang Zihao introduced that reducing the vehicle body weight can lower the load and, under the same battery conditions, strive for longer working time.

The climbing robot displayed by Cainiao

The hardware performance and usage costs of humanoid robots are also issues that have been repeatedly mentioned.

In an exchange with reporters, an industry insider cited dexterous hands as an example, saying that the durability and cost of dexterous hands affect the deployment of humanoid robots. At present, some dexterous hands account for about one-third of the robot's price, and their durability is also less than satisfactory, so in some tasks enterprises are still using gripper solutions.

Ding Xia, taking box handling as an example, said that a dedicated handling device may cost only tens of thousands of yuan and be more efficient, so why would customers use a more expensive humanoid robot to complete the same task? In her view, the grasping ability, stability, heat dissipation and overall cost of humanoid robots still need improvement, and at this stage many applications still struggle to make economic sense.

However, Ding Xia does not deny the future application space of humanoid robots. She believes that as technology matures, output increases and prices fall, some scenarios will gradually become suitable for use. What she cares about more is how many of the quantities announced by enterprises are devices that truly enter commercial use: "Commercialization requires repeat purchases." Responsible editor: Liu Xiuhai. The Paper news tip line: 021-962866. The Paper, unauthorized reproduction prohibited +1 Favorite I want to report #embodied intelligence #industry #logistics #robots View more