您的位置：首页 > 新闻中心 > 社会 > 正文

读心神探

WAIC观察：具身智能“深圳军团”，补齐人形机器人落地关键中间层_我的网站

原来是美男

A | A week ahead of its theatrical release, the trailer of Vaa Vaathiyaar has arrived, and it delivers exactly the mix of oddball humour and unexpected turns that director Nalan Kumarasamy is known for. The film stars Karthi as a police officer whose life is shaped by an intense admiration for the legendary MG Ramachandran, an affection he inherits from his father, played by Rajkiran. The trailer wastes no time establishing this father son dynamic, showing them celebrating MGR with unfiltered devotion and theatre level enthusiasm.The story follows Karthi’s character as he grows up steeped in the mythology of his cinematic hero, only to become a cop whose methods are anything but conventional. His approach to justice often bends rules and rattles superiors, setting the stage for trouble. Nalan frames these beats with his signature mix of sincerity and mischief, giving the audience glimpses of a protagonist who is both heroic and hilariously flawed.A Wild World Filled With Familiar FacesThe trailer also signals the return of Ramesh Thilak and Karunakaran, who previously brought their unpredictable chemistry to Nalan’s cult hit Soodhu Kavvum. Their presence adds instant comedic energy and hints at a story full of twists, confusion and accidental brilliance.The most intriguing reveal belongs to Krithi Shetty, making her Tamil debut. She appears as the protagonist’s love interest, yet the trailer suggests she may exist only inside his imagination. This creative choice mirrors the surreal bend of Soodhu Kavvum, where fantasy and reality drift into each other with a straight face. Her scenes add charm but also mystery, raising questions about how her character fits into the broader narrative.Veteran actor Sathyaraj appears in a political subplot where he and other leaders discuss a brewing public crisis. These glimpses add a sharper edge beneath the comedy, hinting at social commentary wrapped inside absurdity. Anandaraj, GM Sundar and Shilpa Manjunath round out the supporting cast, each bringing a presence that suits the film’s tone of chaos and controlled exaggeration. Craft, Vision and Nalan’s Signature ToneVaa Vaathiyaar’s trailer stands out for its brisk and confident visual design. George C Williams captures the energy of a story that never sits still. Frames move between grounded moments and outright fantasy without losing clarity. The colour palette leans toward vibrancy, matching the theatrical personality of the characters.The film is produced by KE Gnanavel Raja under the Studio Green banner, known for backing projects that combine strong writing with mass appeal. The edit by Vetre Krishnan keeps the trailer tight and rhythmic, while Santhosh Narayanan sets the tone with a score that feels cheeky, charged, and slightly rebellious. DRK Kiran’s art direction builds a world that supports both the realistic and the surreal, a balance that Nalan has always leaned into with confidence.While the trailer keeps major plot beats hidden, it makes one thing clear. Vaa Vaathiyaar is not aiming for a straightforward action comedy. It is pitched as something stranger, funnier and more self-aware. The film blends nostalgic fan culture, unpredictable police drama, political mischief, and moments of pure imagination. In doing so, it feels like a spiritual cousin to Soodhu Kavvum, but with a more heroic central lead and a louder sense of play.Legal Hurdle Clouds December 12 Release PlanJust as the final leg of promotions gathered steam, the film has encountered a serious legal obstacle. The production house behind Vaa Vaathiyaar, Studio Green led by producer KE Gnanavel Raja, faces a lawsuit over unpaid loans. The issue dates back to a 2011 agreement with a financier. The court found that the producer still owed over ₹10.35 crore, and a petition by the official assignee has now resulted in an interim stay on the film’s release. The bench of the Madras High Court has barred the theatrical release until December 5, pending clearance of dues. The next hearing is scheduled for December 8. View this post on Instagram A post shared by Studio Green (@studiogreen_official) This legal roadblock puts the film’s planned December 12 launch in uncertainty. The interim stay affects not just release but the overall financial settlement of the producers. For fans who have been waiting patiently through delays and postponements, this last-minute twist is disheartening. It remains to be seen whether the producers can clear the dues in time or the film’s release will be pushed again.Also Read: Vaa Vaathiyaar Faces Fresh Legal Roadblock Ahead of Scheduled Release。《科创板日报》7月20日讯（记者杨小小）人形机器人要加快实现应用落地，无疑已成为行业内外的普遍共识。围绕这一目标展开的人形机器人“大脑”能力补强，也已经持续了一段时间。但在这个过程中，行业逐渐发现，机器人拥有理解和决策能力，并不等于已经掌握了可执行的技能，从认知到动作之间，还需要完成动作策略生成、本体适配和反馈修正等重要环节。在今年的世界人工智能大会上，具身智能“深圳军团”不约而同地都在朝这个方向进行能力展出，试图缩短模型能力到真机执行的链路。

B | 从数据到动作策略跨维智能将这段距离进一步前移到了模型训练的数据端。此次WAIC，跨维智能发布并开源Aligned DexWorld。据公司介绍，该数据集试图将真人示教、真机遥操和仿真合成等不同来源的数据，在统一的三维物理空间和时间序列中完成对齐。它所针对的是具身智能训练数据的异构问题：不同来源的数据在视角、本体构型和动作空间上存在差异，单纯“堆数据”并不必然提升模型效果，未经对齐的混合训练甚至可能产生负迁移。在数据对齐之外，跨维智能还首次展示了仿真引擎DexVerse的生成式仿真能力，通过持续生成不同的物体姿态、场景布局和交互任务，为世界模型提供更多样化的训练数据。按照公司的技术路径，Aligned DexWorld解决的是不同数据能否被模型共同学习，DexVerse则负责扩展模型可以学习的物理场景。现场展出的手机装盒和商超扫码Demo，进一步呈现了上述技术链路在真机执行端的效果。手机装盒机器人需要连续完成取盒、放机、对位和合盖，并在物料偏移、漏抓等情况下重新规划动作；商超扫码机器人则要适应不同瓶型和随机摆放姿态，完成抓取、扶正、扫码及放置，并通过视觉信号判断扫码是否成功。也就是说，跨维智能试图补上的，是从模型能力到真机执行的上游基础：先让不同来源的数据说同一种物理语言，再让模型学到的能力更顺畅地迁移至真机。如果说跨维智能试图让模型从训练阶段开始，更准确地理解物理世界中的动作，智平方进一步关注的，则是模型理解任务后，如何将认知、协调和执行串成闭环。智平方这次WAIC带来了搭载NeuroVLA类脑大模型的AlphaBot 2。按照公司的设计，NeuroVLA将机器人控制体系分为“皮层-小脑-脊髓”三个层次：皮层负责语义理解和任务规划，小脑负责高频运动协调与动态修正，脊髓则负责毫秒级运动执行和安全反射。高层模型的任务规划无法直接转化为连续的关节动作。机器人还需在执行中持续感知环境和自身状态，在目标变化或动作失败时动态调整，并维持长流程任务的时序信息。

C | 智平方强调的主动感知、故障自恢复和时序记忆，就是为了打通这些环节。在WAIC展示的PCB上下料任务中，AlphaBot 2需要面对不同型号、尺寸的PCB，以及透明、反光托盘，连续完成托盘取料、上下料、移动运输和双臂协同等动作。

D | 相比完成一次固定抓取，这类任务更考验感知、任务规划、动态修正以及运动执行之间能否形成闭环。从动作策略到真机执行模型形成动作策略之后，还需要经过具体场景和机器人本体的适配。乐聚机器人试图通过后训练，补上通用模型与具体岗位技能之间的距离。

E | 作为以本体和运动控制见长的人形机器人企业，在路径上，乐聚没有选择从头训练通用具身基模，而是由上游模型企业完成通用大脑的预训练，再结合自身机器人本体和工业场景开展垂域后训练、本体适配和真机评测。乐聚机器人技术负责人王强对《科创板日报》记者表示，这一过程可以类比为是给机器人开设“技校”：基础模型解决认知广度，后训练则让机器人围绕具体岗位学习技能。以快递分拣为例，蚂蚁灵波的LingBot-VLA 2.0已经通过预训练掌握识别物体、理解指令和基础抓放等能力；乐聚进一步采集300条任务数据进行后训练，让模型学习不同包裹在具体环境中的识别、抓取与分类方法，并将其部署至夸父机器人完成真机任务闭环。据悉，乐聚目前已将数据处理、代码适配、模型后训练和真机部署测试等流程封装为后训练工具链。此次WAIC，乐聚展示的纸箱拆垛、塑料箱拆垛和小件上料，也对应了这套转化路径。乐聚方面人士对记者表示，在展会期间，夸父机器人连续三天每天接近10小时不间断完成了纸箱拆垛等任务，机器稳定性出现了初步的量变突破。

F | 逐际动力则把认知到执行的链路进一步拆分为Sys2认知层、Sys1技能层和Sys0运控层。认知层负责理解任务和作出决策，技能层承接具体能力，运控层则负责将其落实为本体动作。逐际动力此次展示的Oli机器人搭载COSA人形大脑系统和WBT全身运动基础模型。其中，WBT能够实时追踪机器人全身关节，并在动作执行过程中维持平衡。现场动捕演员连续完成深蹲、太极和舞蹈等动作，Oli则进行实时复现。与单个预设动作相比，这一演示更侧重展示长程动作如何被映射至机器人本体并稳定执行。在操作端，逐际动力还展示了搭载COSA系统的TRON2双臂机器人。

G | 现场，TRON2通过视觉感知和语音交互，自主抓取糖果并递送给用户，呈现出机器人从感知、技能生成到底层执行之间的衔接。可以看到，随着人形机器人产业的可见度不断提升，行业内外的期待也在持续拉高。人形机器人企业的展示无疑都在试图回应同一个问题：如何在技术路径尚未完全收敛的情况下，跨过实用性的门槛。但WAIC的演示仍不同于规模化应用。人形机器人面对新任务需要多少数据、训练和适配周期多长，技能能否跨本体、跨场景复用，以及在长期运行中能否保持稳定，仍需更多真实项目验证。