The new humanoid robot from Shanghai startup AgiBot moves like a teenager. With its advanced AI, it’s also emotionally intelligent.
AgiBot, which has already started mass producing humanoids, just introduced the robot named Lingxi X2 alongside its new AI model for enhanced embodied intelligence and human-robot interaction.
In demo footage, the robot rides a scooter, using its foot to propel itself forward, as well as a hoverboard and a self-driving bicycle. Standing 1.3 meters (4 feet 3 inches), the Lingxi joins a growing number of short Chinese humanoid robots like Unitree Robotics’ G1, EngineAI’s PM01, and the T1 from Booster Robotics.
AgiBot, which is also known as Zhiyuan Robotics, says the new humanoid’s hardware system is built with reusable core components. Its Xyber-Edge cerebellar controller enhances movement control and flexibility while its DCU handles high-level motion and intelligent decision-making.
The robot is trained with reinforcement learning using Nvidia’s Isaac Gym Simulator. It’s the same software that Unitree and EngineAI use to achieve human-like bipedal locomotion and fluid movements. With the training method, a robot’s digital twin learns in a virtual environment. The learning is then transferred to the physical robot via a process called Sim2Real, short for Simulation to Reality.
For communication, the robot uses a large language model (LLM) for speech, vision, and gesture-based interactions. It can understand objects, recognize environments, and follow written or verbal instructions. The robot can also be upgraded with dexterous hands for precise object handling.
The humanoid’s generative motion engine, AI-driven interactions, and zero-shot generalization capabilities equip it for a wide range of applications:
- For security and surveillance, the robot can patrol and monitor areas, recognizing faces, analyzing behaviors, and detecting potential threats.
- In healthcare and household settings, it can assist the elderly and disabled with daily tasks, remind patients to take their medication, and support physical therapy with AI-driven assistance
- Its dexterity and social abilities could also prove useful in industrial, retail, hospitality, and public safety use cases
Agibot hasn’t publicly disclosed a price for the new robot. The humanoids in its class start at around $13K for the PM01, $16K for the G1, and $20K to $30K for the T1. Full-sized general-purpose Chinese humanoid robots like the Unitree H1 and Fourier GR-2 cost in the six figures.
AgiBot also just announced what it calls the world’s first universal AI model for humanoid robotics. The Go-1 model, short for Genie Operator-1, uses a Vision Language Latent Action (ViLLA) architecture. Using the model, robots like AgiBot’s Genie-1 and A2 can process images and speech together while predicting and generating actions rather than following preprogrammed movements.
A so-called Mixture of Experts (MoE) uses specialized AI models for different tasks. DeepSeek also uses the MoE technique in its AI models.
With Go-1, robots learn with help from human videos and demonstrations. According to AgiBot, the model adapts across different form factors so robots can share intelligence. It continuously evolves from real-world information to become smarter over time.
AgiBot, which says if manufactured 1,000 humanoids in 2024, also maintains what it calls the world’s largest humanoid manipulation dataset. The data is from a large data collection facility and experimental base in Shanghai.
The facility covers more than 4,000 square meters and contains over 3,000 real objects. It replicates household, industrial, retail, and office environments. The company says it plans to release more than a million real-world trajectories and 10 million digital twin simulation data points in 2025.
AgiBot World was created in collaboration with OpenWorldLab, a research initiative focused on robotics and autonomous driving technologies that’s associated with the University of Hong Kong and the Shanghai AI Lab. The dataset is available for non-commercial use at GitHub and Hugging Face. Any effort to profit from a trained robot would require additional approval or licensing.
Peng Zhihui founded AgiBot in February 2023. The 31 year old previously gained recognition for his work at Huawei developing an Iron Man-inspired robotic arm and a self-driving bike. The founder is positioning AgiBot’s flagship humanoid robot, the A2, as a direct competitor to Tesla’s Optimus.