LimX Dynamics, a Chinese humanoid robotics company, has completed a Series B funding round, raising around USD 200 million.
Institutional investors who took part in this round include UAE-based Stone Venture, Oriental Fortune Capital, Shenzhen Co-Stone Asset Management, Tianjin Venture Capital, GF Xinde Investment, Hefei Innovation Investment, Guotai Junan Innovation Investment, CSC Financial, Tang Xing Capital, and Caixin Capital Investment. Strategic industry investors include JD.com, Zhongding Sealing, NRB Corporation, and Kyland. Existing shareholders Shang Qi Capital, backed by SAIC Motor, along with Bian Era Technology Holdings, Nio Capital, and Future Capital, also increased their stakes.
Embodied intelligence, a sector that has drawn heightened attention over the past two years, is now approaching a critical juncture. After a wave of lab demonstrations and carefully staged showcases, the focus has shifted to a more demanding question: how to deploy robots in real industrial environments, scale delivery, and ensure reliable autonomous operation in a constantly changing physical world. Increasingly, this has become the benchmark by which a robotics company’s technical value is judged.
Accordingly, LimX has introduced two technologies in quick succession: Tron 2, a multi-form embodied robot, and Cosa, an embodied agentic operating system. Tron 2 is positioned as a reusable, scalable, general-purpose base, while Cosa is designed to function as the robot’s cognitive and scheduling hub.
These are not standalone offerings. Working together through Oli, the company’s full-size humanoid robot, they form an integrated solution aimed at addressing a longstanding challenge in robotics: the disconnect between cognition and physical action.
For years, high hardware customization costs and lengthy development cycles have constrained robotics at scale. Under traditional development models, differing task requirements often necessitate building entirely new robots, an approach that has proven costly and inefficient across the industry.
Tron 2 is intended to address this through a full-body modular architecture that enables a single base platform to be reconfigured into multiple forms. Using modular assembly, Tron 2 can shift between dual-arm, bipedal, and wheeled-biped configurations, adapting to tasks that range from fine manipulation to navigation across complex terrain.
In practical terms, the system functions as a set of standardized robotic building blocks, enabling enterprise users and developers to assemble task-specific configurations more quickly and with lower upfront cost.
For tasks requiring precision, such as meal preparation in household settings or sample handling in laboratories, the dual-arm configuration provides humanlike dexterity. Each arm offers seven degrees of freedom, a reach of 70 centimeters, and a payload capacity of up to ten kilograms, sufficient for most common objects.
In environments with uneven terrain, including warehouse staircases or outdoor inspection routes, the bipedal and wheeled-biped configurations provide improved mobility, endurance, and load capacity, supporting up to four hours of operation and payloads of as much as 30 kilograms.
The most immediate benefit of this general-purpose base is cost efficiency. Rather than purchasing dedicated robots for each new application, companies can rely on a single Tron 2 unit and adapt it through modular reconfiguration, improving hardware utilization and return on investment.
Tron 2 also serves as a standardized validation platform. Enterprise customers can use it to test algorithms and commercial workflows earlier in the development process and at lower cost, shifting the most uncertain and expensive trial-and-error phases forward and reducing overall deployment risk.
A flexible physical platform alone, however, is insufficient. General-purpose robots also require system-level intelligence capable of handling varied environments and complex task sequences.
Despite advances in vision-language-action models, many systems remain constrained to linear task execution, in which a single instruction maps to a single action. They often lack persistent environmental awareness and the ability to dynamically plan across competing tasks. LimX introduced Cosa to address this limitation, positioning it as the industry’s first embodied agentic operating system.
In live demonstrations, when the humanoid robot Oli running Cosa receives an instruction such as delivering water to a front desk and is then interrupted with a request to collect parcels, the system does not stall or behave unpredictably. Instead, Cosa’s cognitive layer reassesses task priority and autonomously replans, completing the delivery before retrieving the parcels while coordinating navigation, obstacle avoidance, and manipulation in parallel.
The process runs end to end without relying on predefined scripts. According to LimX, this reflects a shift in system architecture rather than a marginal model improvement. Cosa operates as a central dispatcher, integrating high-level task understanding, mid-level skill libraries, and low-level motion control to enable continuous autonomy from reasoning through execution.
Within this architecture, the modular Tron 2 base, the Oli humanoid robot, and the Cosa operating system, combined with the company’s proprietary foundation model, form a reusable production framework for embodied intelligent agents.
Tron 2 and Oli provide standardized physical platforms designed for stability and precision. A cerebellum model generates real-time full-body motion, while coordination between high-level planning and low-level control systems aligns navigation, obstacle avoidance, manipulation, stair climbing, and other behaviors with whole-body control. Cosa sits at the center, responsible for task comprehension, planning, and orchestration.
This division of labor reduces development friction. Engineers can avoid rebuilding hardware or low-level motion systems from scratch and instead build on LimX’s validated platforms and integrated cognitive-motor stack, using open interfaces and development tools to focus on application layer innovation within specific verticals.
More broadly, this approach represents a shift in LimX’s position within the robotics ecosystem. Rather than competing across every layer, the company is defining clear technical and market boundaries.
As robotic bodies and cognitive systems become increasingly standardized, the company’s challenge moves toward deploying these systems in environments that demand adaptability and close human interaction.
Guided by the principle of serving people rather than production processes, LimX is prioritizing scenarios that require direct interaction and contextual flexibility. Initial deployments are focused on business-facing service environments such as reception and guided tours, with plans to expand gradually into consumer household services.
Internationally, the company plans to pair China’s supply chain advantages in rapid iteration and cost control with localized ecosystems of developers, customers, and distribution partners. The aim is to embed embodied intelligence more deeply into overseas markets.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This article was written by Huang Nan for 36Kr.