作者 | 许丽思
编辑 | 漠影
今年工博会现场，真的太卷了。
400多家机器人企业同场竞技，有的秀抓取，有的拼精度，这三台机器人却靠着组队干活，吸引了不少海外观众驻足围观。
拆垛、倒料、接料、码垛……三台纪灵-L1轮番上阵，接力完成一整套物料流转流程，整个流程看起来非常丝滑，全程无需人工介入！
中科第五纪则把展示推向了更高层：让三台机器人自主接力，把多个工位的活连起来干，以多机协同探索从“单机智能”走向“群体智能”。
这场演示的分量，就在于“接力”二字。三台机器人既要各自完成操作，又要协调物料交接与任务接续。感知、操作、移动和任务调度共同发挥作用，才有了现场一气呵成的连续作业。中科第五纪由此将技术积累转化为看得见的作业能力，让具身智能的真实生产力有了直观呈现。
多机协同、两大新品同步登场，中科第五纪在工博会上亮出了从具身大脑到工业作业的完整实力。
而在展台之外，这家公司已在工业料箱搬运、新能源场站巡操等场景开展落地实践，并获得海外头部企业合计数亿元订单。从现场演示到客户交付，中科第五纪正将模型、产品与协同作业能力，带入更多真实生产场景。
一、从单操作到全流程，工业赛道验收的是完整任务执行能力
以工厂常见的物料转运为例，从取料、搬运到交接、码垛，涉及多个环节：机器人既要按时将物料送到指定工位，避免影响下游生产，还得应对物料偏移、通道受阻、设备异常等突发情况。
这也是去年以来，“群体智能”逐渐成为具身智能工业落地关键词的原因之一。机器人单项操作能力不断提升，多家企业开始探索如何让多台机器人协同作业，将原本分散的操作串联成完整的生产流程。
在工博会现场，中科第五纪展示的三机协同作业，就提供了一个颇具代表性的观察样本。
三台机器人各自能把活干好，只是基础；让整套流程顺畅运转，难度又上了一个台阶。
比如，上游机器人完成拆垛后，下游设备需要及时接收物料；某台机器人操作时间变了，其他设备也得立刻调整执行节奏。单台机器人的操作能力，直接影响整套系统的运行效率。
过去，许多自动化产线围绕固定设备和工序进行设计。多台机器人具备协作能力后，工厂有机会将跨工位物料转运、多环节上下料等任务组织到统一的作业流程中，提高柔性生产的调整空间，使得“智能岛”工作范式在工业生产中成为可能。
据了解，中科第五纪的这套方案瞄准了工业生产中大量重复、高强度且需要连续衔接的任务，面向3C电子、汽车零部件和轻工制造等领域，可应用于精准上下料、箱体物料转运、规整码垛及废料回收等作业，适配多品类产线需求。
众机协作背后，也折射出工业具身智能下一阶段值得关注的竞争变化：随着机器人越来越能干，行业竞争的重点，正在从单机有多强，进一步走向整套流程能创造多少价值。
二、打造跨本体通用的具身大脑，解决能力复用难题
想让机器人真正进入生产流程，要先解决一个基础问题：每增加一种机器人、每进入一个新工位，是否都要重新训练一遍？
这一问题的背后，是工业机器人长期存在的能力复用难题。
不同工厂的物料、工位布局和操作要求各不相同，同一任务在不同机器人上执行，也会受到运动结构、末端工具和控制方式的影响，进而限制着工业机器人的规模化部署。
为此，中科第五纪推出新一代具身大脑纪元灵脑，试图从具身大脑层面提高已有操作能力在不同机器人和任务之间的复用效率，减少重复训练与适配投入。
据了解，纪元灵脑依托自研FAM系列具身大模型，并采用动作光流（Action Flow）技术探索跨本体通用。
这项技术的核心目标，是要让不同机器人之间的能力能够彼此迁移。当工位需要更换不同结构的机器人，或需要将已有任务扩展至其他设备时，就能够复用一部分成熟能力，减少重复训练和开发投入。
比如，一台轮式人形机器人已经熟练掌握抓取和放置操作，后续迁移到其他机器人时，模型有机会利用此前积累的能力，减少从头训练的工作量。
除了跨本体迁移，纪元灵脑还通过热力图技术实现少样本学习，进一步解决新任务如何快速上手的问题。
工业现场存在大量低频、非标及不断变化的任务，逐一为这些任务采集大规模示范数据，往往成本高、周期长。机器人能够通过少样本学习掌握新操作，减少数据采集与训练投入，就能缩短任务导入和现场部署的时间。
跨本体通用与少样本学习结合，就有机会改变工业机器人的部署方式和长期成本结构，共同作用于机器人从单点应用走向多工位、跨产线复制的经济性。
三、工位各有难题，机器人的身体也得各有所长
具身大脑还需要硬件载体，要根据具体任务选择合适的机器人形态，才能将模型能力转化为具体的生产动作。
承担此次众机协同演示的纪灵-L1，采用轮式底盘与双臂结构。轮式底盘满足机器人跨位置移动的需求，双臂则用于抓取、搬运和物料操作，适合移动作业与多机协同。
移动物料流转需要机器人灵活穿梭，固定工位的精密装配则更看重操作范围、精度和工具切换能力。
围绕这些不同需求，中科第五纪此次还发布了立柱式具身智能机器人纪灵-S1。
纪灵-S1具备19个自由度，通过立柱升降与360°腰部旋转，能够覆盖不同高度和方位的操作空间。结合高精度力控和末端模块化快换设计，纪灵-S1适配精密装配、产线作业和智能仓储等场景中的不同工序需求。
从轮式双臂机器人纪灵-L1到立柱式机器人纪灵-S1，中科第五纪围绕不同工业任务布局多种本体，以工业级设计承接工位需求，以众机协同贯通作业环节，推动具身机器人从独立工位走向连续生产流程。
这也反映出工业具身智能的一种发展思路：机器人既需要匹配具体工位的硬件需求，也需要降低不同设备与任务之间的适配成本，才能从单项能力展示走向更广泛的生产应用。
四、落地多个真实场景、手握数亿订单，具身智能商业化开始拼交付
机器人在展台上完成一整套流程，展示了其承担复杂任务的潜力。但真正进入工厂，面对客户的生产要求、现场工况和长期运行考验，技术演示的成功还只是第一步。
在这一过程中，中科第五纪已经积累了多个实际场景的应用经验，并获得了一批产业客户的认可。
目前，其已在工业料箱搬运、智能装配、新能源场站巡操等多个场景已开展落地实践，合作伙伴包括零跑汽车、中国石化、美的等多家业内头部企业。
在电力巡操领域，据此前披露，搭载其具身操作大脑的机器人已在某变电站完成室内外高精度巡操任务，操作成功率超过96%，人工现场作业频次降低70%。这些项目数据，为具身智能能否承担实际工作、减轻现场人员作业负担提供了具体参照。
海外业务也在同步推进。据中科第五纪介绍，目前已获得海外多家知名头部企业合计数亿元订单，相关订单履约与交付持续推进。 接下来将进一步布局欧洲、澳新、日韩等重点市场，推动具身智能机器人产品进入更多海外产业场景。
近日，中科第五纪还与博彦科技新加坡国际总部达成合作。双方计划结合中科第五纪的具身大脑技术，以及博彦科技在东南亚的数字化转型、企业技术服务和本地化交付能力，围绕柔性搬运、灵巧作业与巡操等需求，推进机器人产品及定制化解决方案落地。
从真实场景验证到获得海外订单，中科第五纪正在加快将技术能力转化为能够面向产业客户交付的解决方案。
结语：具身大脑的长期价值，在于能力复利
中科第五纪已围绕这一目标形成了清晰布局：纪灵-L1与纪灵-S1面向不同工业任务，纪元灵脑则尝试连接不同本体、任务，其在跨本体迁移、少样本学习上的验证进展，也为能力复用提供了初步基础，也为机器人更快适应新任务、在更多工业场景中创造真实生产力提供了支撑。
更值得关注的是，随着真实部署增加，机器人积累的数据与操作经验，有望持续反哺模型迭代，降低后续项目的重复开发成本。
工业具身智能的下一轮竞争，拼的将是能力能否持续积累、成本能否持续下降。让每一次交付都成为下一次部署的技术积累，才有机会真正打破重复开发的循环，打开规模化落地的空间。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
I'm the author.
Edit |
This year's Expo is really too much.
More than 400 robotics competed, some were captured, and some were refined, while the three robots worked in teams, attracting many viewers from abroad.
Dismantling, refilling, relaying, coding... three gills-L1 round, relaying a full material flow process that looks very silky, without manual intervention!
Middle Queens V pushed the show to a higher level: to connect the three robots, to connect the work of multiple jobs and to explore multiple opportunities to move from “one- unit intelligence” to “group intelligence”.
The point of this demonstration is the word "relay". The three robots are required to perform their own operations while coordinating the handover of materials and the succession of tasks. The synergy of perception, operation, movement and task control has enabled continuous, on-site operation. As a result, mid-level quinquies transformed technological accumulation into visible operational capability, giving a visual look at real productivity with intelligence.
Multi-channel and two new products are coming in handy, and the full power of China's fifth century emerged at the Expo, from the brain to the industrial work.
Outside the exhibition stand, the company has undertaken landing practices in such settings as industrial tank handling and new energy station patrols, and has received orders for several hundred million yuan from head companies overseas. From the field demonstration to the customer delivery, Middle East Section V is bringing models, products and co-operation capabilities to more real production scenes.
I. From single operation to full process, industrial track acceptance check is fully operational.
In the case of trans-shipment of materials, which is common in factories, from material extraction, handling to handing over, coded, involves multiple links: The robot has to respond to sudden material deviations, obstructions of access and equipment anomalies, as well as to the timely delivery of materials to designated posts to avoid affecting downstream production.
This is one of the reasons why, since last year, “group intelligence” has become a key word for smart industry. The robotics' individual operating capabilities are constantly increasing, and many enterprises are beginning to explore ways in which multiple robots can work in tandem, linking previously dispersed operations into complete production processes.
At the Expo site, a representative observation of sample was provided by the concerted operation of the three machines displayed by Middle East and East Asia.
Each of the three robots can do the job well, but it's just the foundation; it's a step up to keep the whole process running.
For example, when the upper robot is finished, downstream equipment needs to receive the material in a timely manner; one robot has changed operating time and others have to adjust the execution rhythm immediately. The operational capability of a single robot directly affects the operational efficiency of the entire system.
In the past, many automated production lines were designed around stationary equipment and processes. With the collaborative capability of multiple robots, plants have the opportunity to organize cross-site trans-shipment of materials, multi-stage substrates, etc. into a single operating process, and to increase the adjustment space for flexible production, making the “smart island” work paradigm possible in industrial production.
It is understood that this set of programmes, aimed at high levels of duplication, intensity and continuous interconnectivity in industrial production, can be applied to such areas as 3C electronics, auto parts and light manufacturing, such as precision downloading, transhipment of box materials, coded stacks and waste recycling, and can be adapted to multiple product line requirements.
The next stage of competitive change in industrial intelligence is also reflected in the collaboration of the public: As robots become more capable and industry competes, they are moving from the strength of a single machine to the value that the whole process can generate.
II. Building a human brain that is universal across the body to solve the problem of re-use of capacity
In order for robots to truly enter the production process, a fundamental question needs to be addressed: is there a need to retrain every additional robot that enters a new position?
Behind this problem is the challenge of the long-term re-use of industrial robots.
Material, site layout and operating requirements vary from plant to plant, and the same task is carried out on different robots and can be influenced by motor structures, end-of-pipe tools and controls, thus limiting the deployment of industrial robots to scale.
To this end, the Fifth Century of China has introduced a new generation of brain epochs that seek to increase the efficiency of existing operational capabilities at the brain level between different robots and missions and to reduce duplication of training and adaptation inputs.
It is understood that the circulatory brain relies on self-study of the FAM series, with its body LLM, and uses action Flow technology to explore cross-introvert generality.
The core objective of this technology is to enable the migration of capabilities between different robots. When jobs need to be replaced with robots with different structures, or where existing tasks need to be extended to other equipment, a part of the mature capacity can be reused to reduce duplication of training and development inputs.
For example, when a wheeled human robot is already skilled in capture and placement operations and is subsequently relocated to other robots, the model has the opportunity to take advantage of previously accumulated capabilities to reduce the volume of training from scratch.
In addition to trans-incorporate migration, the epoch further addresses the quick start of new tasks through the realization of less sample learning through the hot pursuit of technology.
Industrial sites have a large number of low-frequency, non-standard and changing missions, and large-scale demonstration data are collected for each of these missions, often at high cost and long cycles. The ability of robots to learn new operations by learning less sample and to reduce input for data collection and training will reduce the time spent on mission importation and on-site deployment.
The integration of cross-intellectual interoperability with less sample learning offers an opportunity to change the way industrial robots are deployed and the long-term cost structure, working together on the economics of robots moving from single-point application to multi-position, cross-production line replication.
iii. Each job has a problem and each robot's body has to be good.
A physical brain also requires a hardware carrier that selects the appropriate robotic form according to the specific mission in order to translate model capabilities into specific production actions.
The Grinch-L1, which carries out the crowd demonstration, uses a wheeled chassis and arms structure. The wheel-based chassis meets the need for robots to move across positions, while the arms are used for capture, handling and material operations, which are suitable for moving operations in synergy with multiple machines.
Mobile material flow requires robotic flexibility to travel, and the precision assembly of fixed positions is more focused on the range of operations, precision and tool switching.
In response to these different needs, publish has also been set up this time by ZMC5.
KIL-S1 has 19 degrees of freedom and rotates from the height of the pole to the waist at 360°, covering operating space at different heights and locations. Different process requirements in scenarios such as high-precision control and end-end modularization fast-change design, gill-S1 fit-in precision assembly, linework and intelligent warehousing.
From the wheeled double-arm robot kirin-L1 to the vertical robot irix-S1, the fifth century of China has been designed around a variety of different industrial tasks, designed to take over the workspace requirements at the industrial level, and to connect people to the operational chain, facilitating the movement of robots from stand-alone positions to continuous production processes.
It also reflects an industry-smart approach to development: robots need to match the hardware needs of specific jobs with lower matching costs between different equipment and tasks in order to move from individual capabilities to broader production applications.
iv. Multiple real scenes, hundreds of millions of orders at hand, smart commercialization begins to deliver
The robot has completed a set of processes on the stage, demonstrating its potential for complex tasks. However, the success of the demonstration is only a first step in the face of the production requirements of the client, the situation on the ground and the test of long-term operation.
In the course of this process, MC5 has accumulated many practical experiences and received the approval of a group of industry clients.
At present, it has been set up in a number of settings, such as industrial tank handling, smart assembly, and new energy station patrols, with partners such as Zero Runners, Chinese petrochemicals and American leading firms.
In the field of electrical surveying, it was previously disclosed that robots carrying their body-operated brain had completed high-precision indoor and outdoor patrols at one of the transformers, operating over 96 per cent of success rate and reducing the frequency of manual on-site operations by 70 per cent. These project data provide specific references to the ability of smart people to undertake actual work and to reduce the operational burden on people on the ground.
Overseas operations are also advancing simultaneously. According to the fifth chronology presentation, orders totalling hundreds of millions of yuan have been obtained from a number of prominent overseas head companies, and the performance and delivery of the related orders have continued to advance. This will be followed by a further deployment of key European, Austro-American, and Japanese and South Korean markets to facilitate the entry of smart robots into more overseas industries.
In recent days, China has also worked with the International Headquarters for Science and Technology Singapore. The two sides plan to promote the landing of robotic products and customized solutions around the needs of flexible handling, smart operations and rotations, in combination with mid-level fifth-century brain technology and the digital transformation of Bohyan technology in South-East Asia, enterprise technical services and local delivery capacity.
From real scene validation to obtaining overseas orders, CS5 is accelerating the transformation of technological capabilities into solutions that can be delivered to industry clients.
Conclusion: the long-term value of having a brain is the ability to recover.
The fifth chronology has developed a clear layout around this goal: Qirin-L1 and Qirin-S1 are oriented towards different industrial missions, while the Qur'an brain tries to connect different ingenuities and missions, whose learning progress across the introsities, less sample, also provides an initial basis for capacity re-use and supports the faster adaptation of robots to new tasks and the creation of real productivity in more industrial scenarios.
Of even greater concern is the fact that as real deployment increases, the data and operational experience accumulated by robots are expected to continue to evolve and reduce the double development costs of subsequent projects.
The next round of smart industrial competition will be about the sustainability of capacity accumulation and the sustainability of cost reductions. By making each delivery the next technological build-up for deployment, there is a real opportunity to break the cycle of redevelopment and open up space for scale.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.