news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

Author: Huang Nan

Editor: Yuan Silai

On August 18, at the 2024 new product launch conference of Zhiyuan Robotics, Zhihuijun hosted and released two major series of "Expedition" and "Lingxi", a total of 5 new commercial humanoid robots.

Aiming at different application scenarios, Zhiyuan’s “Expedition” series has launched the Expedition A2, Expedition A2-W and Expedition A2-Max.

Yuanzheng A2 is an interactive service robot, 1.7 meters tall and weighing 70 kilograms. At the press conference, Yuanzheng A2 appeared as a host, with fluent explanation skills and stable movement functions.

The Expedition A2-W is a flexible intelligent manufacturing robot that adopts a dual-arm + wheel structure, which can achieve fast and stable movement in a smooth environment. In the opening scene of the press conference, the Expedition A2-W demonstrated its operational capabilities such as dynamic task scheduling, complex operation execution and dual-arm collaborative operation by making drinks for Zhihuijun.

The Expedition A2-Max is a heavy-load special robot with the advantages of immense strength and dexterity. It can move 40kg aviation boxes and is currently in the product development stage.

The two robots in the “Lingxi” series are from Zhihuijun’s laboratory X-Lab and were completed by a team of 10 people in less than 3 months.

Among them, Lingxi X1, as a full-stack open source robot, can support "machine-machine mode", adopts modular design, has the characteristics of high degree of freedom, safety, light weight and high scalability, and most importantly, it is priced at 0 yuan. Zhihuijun said that the main body design drawings, software framework, middleware source code, basic operation and control algorithms of Lingxi X1 will be open source, and core components will be sold. It is hoped that in this way, the arrival of the era of "humanoid robots made by humans" will be promoted.

Lingxi X1-W is a professional data acquisition robot that only collects real data and serves the pre-training stage to reduce data acquisition costs.

Jiang Qingsong, partner and vice president of marketing services of Zhiyuan Robotics, revealed,In 2024, Zhiyuan Robotics estimates that its shipments will reach around 300 units.Among them, the humanoid robot is expected to start mass production in October, with a shipment volume of about 200 units, and the wheeled robot is expected to ship 100 units.

Zhiyuan Robot Family Portrait

Specifically in terms of technology, Zhiyuan divides the robot system into power domain, perception domain, communication domain, and control domain.

In the power domain, Zhiyuan's self-developed PowerFlow joint module has achieved mass production and iterative upgrades. The rated torque of the strongest model reaches 270Nm and the peak torque reaches 512Nm. The number of degrees of freedom of the dexterous hand has jumped to 19, and the active degrees of freedom have doubled to 12. In order to enhance the perception ability, tactile perception and visual tactile perception technology based on MEMS principles have been introduced.

In terms of perception domain, Zhiyuan has integrated sensors such as RGBD cameras, lidar, panoramic cameras, etc. in the new generation of robots, introduced cutting-edge perception solutions for autonomous driving occupancy, and further improved environmental understanding capabilities through SLAM algorithms.

In the communication domain, the self-developed AimRT, a native, lightweight, high-performance intelligent robot communication framework, has improved performance, stability, system deployment efficiency and flexibility compared to third-party middleware such as ROS, while being fully compatible with the existing ROS/ROS2 ecosystem. AimRT will be open source at the end of September.

In the control domain, Zhiyuan combines the Model-based and Learning-based algorithms to further improve the robot's motion control and adaptability. It has also pre-developed AgentOS, which is driven by a natural language instruction set and can adapt to different robot bodies. Based on reinforcement learning, it can achieve precise orchestration and efficient execution of robot skills.

In addition, Zhiyuan has also built an open ecosystem-oriented software platform AIMA (AI Machine Architecture), covering on-machine, cloud and client. This platform not only integrates the core technology of robot software, but also provides a rich secondary development interface.

Four-domain layout

At the press conference, Zhiyuan proposed the evolution route of embodied intelligent technology from G1 to G5, namely: G1 basic automation, G2 general atomic skills, G3 end-to-end operation skills, G4 end-to-end operation large model, and G5 AGI, a total of five routes.

In the past year, Zhiyuan has made a breakthrough in the G2 route, and has realized a series of zero-shot and few-shot general atomic skills, including the universal pose estimation model UniPose, the universal grasping model UniGrasp, and the universal force control plug-in model UniPlug. The atomic capability model in the G2 stage is aimed at flexible intelligent manufacturing and interactive service scenarios and can be commercially applied in multiple practical scenarios.

On the G3 route, Agibot has developed a complete full-process embodied data solution AIDEA (Agibot Integrated Data-system for Embodied AI). In the fourth quarter of this year, one million real machine data and ten million simulation data based on AIDEA will be open source.

The evolution of embodied intelligence technology from G1 to G5

At present, Zhiyuan has cooperated with companies such as Junpu Intelligence, Digital China, iFlytek, Nortel Digital Intelligence, and Neusoft to create commercial solutions for different scenarios; at the same time, it has cooperated with top scientific research institutions such as the Shanghai Artificial Intelligence Laboratory and the Institute of Software of the Chinese Academy of Sciences, focusing on the technical research and development of multimodal large models and robot operating systems.