LimX Oli

60 000 USD

Limx Dynamics
Full-size (165 cm) humanoid for embodied-AI R&D with a 31-DoF body, dual depth cameras, Python-friendly SDK and simulator support — scalable to higher-DoF, dexterous configurations.
Skill Score
3Specifications and details:
| Availability | In production |
|---|---|
| Nationality | China |
| Website | https://www.limxdynamics.com/ |
| Degrees of freedom, overall | 43 |
| Degrees of freedom, hands | 10 |
| Height [cm] | 165 |
| Manipulation performance | 3 |
| Navigation performance | 3 |
| Max speed (km/h) | 5 |
| Strength [kg] | 3 |
| Weight [kg] | 55 |
| Runtime pr charge (hours) | 1.5 |
| Safe with humans | Yes |
| CPU/GPU | RK3588/8G/64G |
| Ingress protection | Yes |
| Camera resolution | Intel® RealSense™ D435i |
| Connectivity | WiFi 6/USB3.0&3.2/Gigabit Ethernet (RJ45)/Type-C |
| Operating system | Linux |
| LLM integration | Yes |
| Latency glass to action | 1–10 ms |
| Motor tech | Not specified |
| Gear tech | Not specified |
| Main structural material | Not specified |
| Number of fingers | 10 |
| Main market | Global |
| Verified | Not verified |
| Walking Speed [km/h] | 3 |
| Shipping Size | N/A |
| Color | N/A |
| Manufacturer | Limx Dynamics |
Description
Update · July 2026 – COSA 0.5: autonomous household chores
In July 2026, LimX made Oli the showcase for COSA 0.5, the latest version of its three-layer «Humanoid Brain System»: an embodied agentic OS with memory and a world model (System 2), VLA skill models trained through the open-source FluxVLA Engine (System 1), and whole-body motion control (System 0). In the accompanying demo, Oli works through a full room of household chores in one continuous take — with no teleoperation and no cuts, according to LimX — which the company calls China’s first fully autonomous long-horizon mobile manipulation in a home by a full-size humanoid. A full technical report is announced. Read our coverage of the demo and LimX’s $200M Pre-IPO round
LimX Oli is a full-size, general-purpose humanoid built to move — and learn — like a person. It stands 165 cm tall at around 55 kg and uses a 31-DoF body to deliver smooth, human-like motion for research, prototyping, and public demos. Multiple end-effectors are supported so teams can swap hands or grippers as projects evolve.

I’m Olivia, Humanoid Analyst at Humanoid.Guide
My mission, together with the team, is to help your organization understand the landscape, compare solutions, and move toward successful robot deployments.
For perception, Oli pairs a self-developed 6-axis IMU with head- and chest-mounted Intel RealSense D435i depth cameras, giving it a dense view of the environment. Integration is straightforward thanks to USB 3.0/3.2, Gigabit Ethernet (RJ45) and onboard 24 V/12 V power for adding LiDAR or extra cameras. A modular design allows fast assembly and maintenance.
Developers get a modular SDK with high-level and low-level motion APIs, full Python support, and OTA-updated motion libraries. A “zero-gap” URDF enables Sim2Real workflows across major simulators like NVIDIA Isaac Sim, MuJoCo, and Gazebo, backed by documentation and examples.
Oli ships in Lite, EDU, and Super trims. Depending on configuration, total active DoF can scale up to 43 (with dexterous five-finger hands), and compute options include RK3588 for motion control plus NVIDIA Orin NX for perception, with an optional AGX Orin backpack on higher trims. Specifications vary by version.











