CtrlK
| LimX CL-1 |
|---|
| General information |
| Manufacturer |
| Country of origin |
| Year unveiled |
| Status |
| Robot type |
| Height |
| Degrees of freedom |
| Max speed |
| Locomotion |
| Sensors |
| Actuators |
| Website |
The LimX CL-1 is a bipedal humanoid robot developed by LimX Dynamics (formally registered as Nanke Xiaobai Robot Technology (Shenzhen) Co., Ltd.), a Chinese robotics company headquartered in Shenzhen. Unveiled in December 2023, the CL-1 was the company's first full humanoid prototype and attracted widespread attention for its ability to dynamically climb stairs using real-time terrain perception, a feat that only a handful of humanoid platforms worldwide had demonstrated at the time. The robot served as the foundational testbed for LimX Dynamics' approach to embodied AI, combining proprietary actuators, model predictive control, and reinforcement learning (RL) to achieve robust locomotion across varied terrain.
The CL-1 launched a rapid iterative cycle that produced the CL-2 and CL-3 variants before the company consolidated its humanoid platform into the commercially available LimX Oli robot in mid-2025. Together, the CL series established LimX Dynamics as one of the most closely watched humanoid robotics startups in China.
LimX Dynamics was founded in 2022 by Wei Zhang, a tenured professor at the Southern University of Science and Technology (SUSTech) in Shenzhen. Zhang holds a Ph.D. in electrical and computing engineering from Purdue University, completed in 2009, and previously held a postdoctoral position at the University of California, Berkeley, as well as a tenured faculty role at Ohio State University before joining SUSTech in 2019.[1][2] Zhang's academic research focused on motion control, optimization, and legged locomotion, areas that directly informed the technical direction of LimX Dynamics.
The company's co-founder and Chief Operating Officer is Li Zhang (Zhang Li), who previously served at WeRide, an autonomous vehicle startup. Li Zhang's experience in the commercialization of autonomous driving technology brought operational and business development expertise to complement Wei Zhang's research background.[3]
LimX Dynamics has raised a total of approximately $296 million across multiple funding rounds, making it one of the best-funded humanoid robotics startups globally.
| Round | Date | Amount | Lead investors | Notable participants |
|---|---|---|---|---|
| Angel | October 2023 | Part of CNY 200M (~$27.4M combined) | FreeS Fund | Stalagnate Capital, Future Capital, Kinzon Capital |
| Pre-A | October 2023 | Part of CNY 200M (~$27.4M combined) | Vitalbridge Capital | Lenovo Capital and Incubator Group |
| Strategic investment | May 2024 | Undisclosed | Alibaba Group (Haoyue Enterprise Management) | Alibaba held ~18.7% stake |
| Series A | July 2024 | Undisclosed | China Merchants Venture Capital, Shang Qi Capital (SAIC Motor) | FreeS Fund, Vitalbridge, Future Capital |
| Series B | February 2026 | $200 million | Lestone Capital (Abu Dhabi), Oriental Fortune Capital | JD.com, Zhongding Group, SAIC Motor, NIO Capital, Stone Venture, Shenzhen Co-Stone, Tianjin Venture Capital, GF Xinde, CSC Financial |
The Series B round, announced on February 2, 2026, was one of the largest single rounds raised by a Chinese humanoid robotics company. The funds were earmarked for scaling manufacturing, expanding the COSA operating system, and supporting global market entry.[4][5]
Beyond the CL humanoid series, LimX Dynamics develops a range of legged and multi-form robots:
Prior to the CL-1, LimX Dynamics had focused on quadruped and biped robot platforms, particularly the W1 wheeled quadruped. The CL-1 represented the company's first venture into full humanoid form factors, combining upper and lower body structures with arms capable of basic manipulation tasks. The robot was designed as a development and testing platform for embodied AI rather than as a commercial end product.
The CL-1 was unveiled at the end of December 2023, accompanied by a demonstration video that showed the robot climbing stairs, walking down slopes, and navigating curbs in both indoor and outdoor environments. The video attracted significant attention online, particularly because the stair-climbing behavior was performed dynamically with real-time terrain assessment rather than pre-programmed step sequences.[11]
The CL-1 stands approximately 157 cm (5 ft 2 in) tall, placing it in a compact humanoid class suitable for navigating standard indoor environments including doorways and corridors. The base skeleton features over 20 degrees of freedom, providing sufficient articulation for bipedal walking, stair climbing, slope traversal, and basic upper-body manipulation.
The robot's actuators are proprietary hollow direct-drive motors developed in-house by LimX Dynamics. These motors prioritize impact resistance and precise torque control while maintaining low weight. The hollow design allows for a more compact joint structure, which contributes to the robot's relatively light build compared to humanoids of similar height. LimX Dynamics has cited the actuator design as a key competitive advantage, enabling capabilities such as deep squats, slope walking, and push-recovery maneuvers at speeds up to 4 km/h.[12]
For perception, the CL-1 integrates a multi-sensor fusion system combining vision cameras and LiDAR. This sensor suite generates 3D maps of the robot's surroundings in real time, feeding data into the gait planning and locomotion control systems. The integration of perception with locomotion, sometimes referred to as "closing the loop," was a central design goal for the CL-1.
The CL-1's control system operates as a closed-loop pipeline connecting four stages: real-time perception, gait planning, locomotion control, and hardware actuation. At each step, the system computes optimal foot landing points, interaction forces, and movement velocities to achieve stable interaction between the robot and the terrain.
LimX Dynamics employs a combination of model predictive control (MPC) and reinforcement learning (RL). The MPC component handles real-time trajectory optimization, while RL-trained policies manage higher-level locomotion behaviors. The RL approach follows a sim-to-real transfer paradigm: policies are first trained in physics simulation (with LimX Dynamics using NVIDIA Isaac Sim), where the robot effectively teaches itself to walk and navigate obstacles through millions of trial-and-error iterations, before being deployed on the physical hardware.[13]
The company also developed a real-time, distributed operating system for the robot with three components: an integrated development toolchain to reduce coding complexity, real-time data transmission for precise robot operation, and decentralized node operation that balances computing loads across the system's processors.
The CL-1's debut video showed the robot performing dynamic stair climbing using real-time terrain perception. Unlike many humanoid demonstrations at the time, which relied on pre-mapped environments or scripted movements, the CL-1 independently assessed each staircase, slope, and curb as it encountered them and adjusted its gait accordingly. In the initial December 2023 demonstration, the robot used a double-step approach, placing both feet on each stair before ascending to the next. The robot also demonstrated walking down a 15-degree slope and stepping onto curbs in diminishing daylight conditions.[14]
LimX Dynamics described the CL-1 as "one of the few humanoid robots around the world that achieves dynamic stair climbing based on real-time terrain perception," attributing the capability to its motion control algorithms, proprietary actuators, and integrated hardware system.[15]
In April 2024, LimX Dynamics released an update showing two major capability advances. First, the CL-1 had progressed from double-step stair climbing to single-stride ascending, alternating feet with each step to mirror the natural human stair-climbing gait. This represented a significantly more challenging control problem due to the longer stance phase and greater center-of-mass displacement on each step.
Second, the CL-1 demonstrated dynamic running for the first time, executing transitions from a standstill to a full sprint and back to a stop. Running required solving several technical challenges, including managing the flight phase where both feet leave the ground simultaneously, maintaining full-body coordination during rapid center-of-mass shifts, and executing smooth speed transitions. The company noted that hardware improvements, including enhanced joint torque, faster rotational speeds, and reduced actuator reaction times, were necessary to support running. Structural refinements also achieved stronger impact resistance at a lighter overall weight.[16]
In August 2024, LimX Dynamics released a three-minute single-take video showing the CL-1 performing warehouse logistics tasks. The robot picked up and carried bins weighing up to approximately 8 kg (18 lbs) from tables and placed them on upper and lower shelves. The demonstration highlighted several capabilities:
This demonstration was notable for being presented as an unedited single continuous take, intended to show authentic real-time performance rather than edited highlights.
LimX Dynamics iterated rapidly on the CL platform through 2024, producing two additional variants before transitioning to the Oli commercial platform.
The CL-2 was announced in 2024 as a next-generation evolution of the CL-1, with particular emphasis on waist and hip actuation. The hip joint design drew comparisons to Boston Dynamics' Atlas robot, though the CL-2's legs lacked the 360-degree rotation capability that Atlas achieves from below the hips. The CL-2 introduced new high-torque electric actuators and a redesigned hollow actuator architecture, enabling a wide range of full-body movements that its predecessor could not achieve.
Demonstration videos showed the CL-2 performing 360-degree torso rotation, deep parallel squats, lying down and lifting itself to standing by rotating its legs, and waist-twisting movements that exceeded the range of human hip joints. LimX Dynamics described the CL-2 as "more compact, powerful, and high-performance" relative to the CL-1.
The company stated that the CL-2 prioritized energy efficiency, using the shortest and most direct movement paths to complete tasks. This design philosophy prioritized function over strict biological mimicry, and LimX Dynamics claimed the robot consumed less energy than comparable platforms such as Atlas and Unitree's H1.[19]
The CL-3 represented the most advanced iteration of the CL series before the platform was commercialized as Oli. Standing approximately 164 cm (5 ft 5 in) tall with a weight of about 45 kg, the CL-3 featured 31 active degrees of freedom in its main body (43 total including 12 DoF in its optional dexterous hands). The joint layout included 7 DoF per arm, 6 DoF per leg, a 3 DoF waist, and a 2 DoF neck.
The CL-3 introduced several new capabilities. In a viral demonstration video, the robot performed a series of human-like stretching and warm-up movements, including arm raises, head rotations, torso twists, side bends, and arm curls. LimX Dynamics also introduced its VideoGenMotion framework with the CL-3, allowing the robot to imitate human actions directly from video inputs, facilitating loco-manipulation tasks such as shelf picking or assembly without extensive manual programming.
The robot's battery provided 2 to 3 hours of operation, and it moved at approximately 1.5 meters per second. The CL-3 was offered to university labs and pilot factories at a prototype price of approximately $40,000.[20]
| Specification | CL-1 (Dec 2023) | CL-2 (2024) | CL-3 (2024) | LimX Oli (Jul 2025) |
|---|---|---|---|---|
| Height | ~157 cm | Not disclosed | ~164 cm | 165 cm |
| Weight | Not disclosed | Not disclosed | ~45 kg | ~55 kg |
| Active DoF | 20+ | Not disclosed | 31 (body) + 12 (hands) | 31 (excl. end effectors) |
| Max speed | ~4 km/h | Not disclosed | ~1.5 m/s (~5.4 km/h) | Not disclosed |
| Battery life | Not disclosed | Not disclosed | 2-3 hours | Not disclosed |
| End effectors | Paddle grippers | Not disclosed | Optional 5-finger hands | 2-finger grippers or 5-finger hands |
| Compute | Not disclosed | Not disclosed | Not disclosed | RK3588 (motion) + NVIDIA Orin NX (perception); optional AGX Orin |
| Sensors | Vision + LiDAR | Multi-sensor fusion | Integrated sensors + AI | IMU, Intel RealSense, industrial camera |
| Price | Not available | Not available | ~$40,000 (prototype) | From RMB 158,000 (~$21,800) |
| Status | Prototype | Prototype | Prototype | Commercial (3 versions: Lite, EDU, Super) |
The CL-3 served as the direct precursor to the LimX Oli, which was officially launched on July 30, 2025, and made its public debut on August 8, 2025, at the World Robot Conference (WRC) in Beijing. Oli retained the 31-DoF architecture and 165 cm height of the CL-3 but increased in weight to approximately 55 kg, likely due to additional structural reinforcement and a more robust compute stack.
Oli was designed as a general-purpose research and integration platform for AI researchers, robotics developers, and solution integrators. It shipped in three versions: Lite, EDU, and Super, with prices starting at RMB 158,000 (approximately $21,800). Key features included a modular hardware-software architecture, an open SDK with access to full-body state and joint-level control, over-the-air (OTA) updates for motion libraries, and support for third-party peripherals including microphones, cameras, tactile sensors, and LiDAR. The sensor system featured a six-axis IMU developed by LimX, an Intel RealSense camera, and a high-resolution industrial camera. Compute options ranged from RK3588 for motion control plus NVIDIA Orin NX for perception on the base model, with an optional NVIDIA AGX Orin backpack available on higher-tier configurations.[9][21]
The transition from the CL series to Oli marked LimX Dynamics' shift from pure research prototyping to commercial product delivery, part of a broader strategy the company calls its IDS (Innovators, Developers, System Integrators) ecosystem.
On January 12, 2026, LimX Dynamics launched COSA (Cognitive OS of Agents), which the company described as the first operating system built from the ground up for humanoid robots operating in real-world environments. COSA was designed as an embodied agentic operating system native to the physical world.
COSA introduced two primary capabilities. The first is Semantic Memory, which allows robots to retain and build upon information about their environment across interactions, rather than treating each task as a fresh encounter. The second is coordinated whole-body control, enabling simultaneous manipulation and locomotion so that the robot can walk and work with its hands at the same time rather than performing these actions sequentially.[22]
The COSA system was subsequently integrated into the Luna humanoid, a performance-focused evolution of the Oli platform.
LimX Dynamics' approach to locomotion control centers on reinforcement learning, an area of machine learning in which an agent learns optimal behavior through trial and error in a simulated environment. The company's RL pipeline follows a sim-to-real methodology:
This approach allows the CL-1 and its successors to develop generalizable locomotion skills rather than terrain-specific routines. The RL-trained policies handle tasks such as gait selection, foot placement, balance recovery from pushes, and speed adjustment. Combined with the model predictive control layer that handles trajectory optimization, the system can adapt to novel environments it has not specifically trained on.
The P1 biped robot served as an important upstream platform for developing and validating RL-based locomotion algorithms before they were applied to the more complex humanoid CL series. Lessons learned from the P1's mountain-traversal experiments directly informed the CL-1's terrain perception and gait planning systems.[7]
LimX Dynamics has identified several primary application areas for its humanoid robots:
The CL-1 arrived during a period of intense global activity in humanoid robotics. In 2023 and 2024, companies including Tesla (with Optimus), Figure AI (with Figure 01 and 02), Unitree (with H1 and G1), Agility Robotics (with Digit), and Boston Dynamics (with the electric Atlas) were all advancing humanoid platforms. Within China, firms such as UBTECH, Agibot, and Fourier Intelligence were pursuing parallel development efforts.
The CL-1 distinguished itself in several ways. Its real-time terrain perception for stair climbing was a genuinely rare capability among humanoid robots at the time of its December 2023 debut. The rapid iteration from CL-1 through CL-3 to a commercial product (Oli) within roughly 18 months demonstrated an unusually fast development cycle. The company's RL-based locomotion approach, validated first on the P1 biped and W1 quadruped before being applied to humanoids, provided a methodical technology transfer pipeline.
LimX Dynamics' $296 million in total funding, capped by the $200 million Series B in February 2026, placed it among the most heavily capitalized humanoid robotics startups worldwide, reflecting investor confidence in its technical approach and commercialization strategy.