CtrlK
| AgiBot RAISE A1 | |
|---|---|
 | |
| General information | |
| Manufacturer | AgiBot (Zhiyuan Robotics) |
| Chinese name | 远征A1 (Yuanzheng A1) |
| Country of origin | China |
| Year introduced | August 18, 2023 |
| Status | Prototype / Technology demonstrator |
| Target price | Below RMB 200,000 (~$30,000 USD) |
| Website | agibot.com |
The AgiBot RAISE A1 (Chinese: 远征A1, Yuanzheng A1, literally "Expedition A1") is a full-size bipedal humanoid robot developed by AgiBot (formerly known as Zhiyuan Robotics), a Chinese robotics company headquartered in Shanghai. Unveiled on August 18, 2023, it was the company's inaugural robot product, arriving just six months after AgiBot's founding in February 2023.[1][2] The RAISE A1 served as a technology demonstrator and industrial research platform, targeting applications such as bolt tightening, vehicle inspections, and laboratory experiments in the automotive, 3C manufacturing (computers, communications, consumer electronics), and logistics sectors.[3]
Standing 175 cm tall with 49+ degrees of freedom, the RAISE A1 introduced several proprietary technologies that would become hallmarks of AgiBot's product line, including the PowerFlow quasi-direct-drive joint actuator, the SkillHand dexterous manipulation system, the EI-Brain (Embodied Intelligent Brain) hierarchical AI architecture, and the WorkGPT multimodal large language model.[4] Although the RAISE A1 itself was never mass-produced, it validated AgiBot's core engineering approach and directly informed the design of the commercially oriented AgiBot A2 series, which launched in August 2024 and went on to become a mass-production product line.[5]
AgiBot was founded in February 2023 by Peng Zhihui and Deng Taihua, both former executives at Huawei. Peng Zhihui, born in 1993 in Ji'an, Jiangxi Province, studied biomedical engineering at the University of Electronic Science and Technology of China (UESTC), where he won over 40 engineering competition awards. He gained a massive online following on Bilibili through DIY robotics projects, including a robotic arm inspired by Iron Man's "Dummy" that garnered 2.5 million views within a day of posting. In 2020, Peng joined Huawei through its highly selective "Genius Youth" (Top Minds) recruitment program, earning an annual salary of 2 million yuan as an AI algorithm engineer. He departed Huawei in December 2022 to pursue robotics entrepreneurship.[6][7]
Deng Taihua brought 27 years of Huawei experience to the venture. He had served as Vice President and President of the Computing Product Line, overseeing ecosystem development for the Kunpeng Processor and Ascend AI Chip. Together, Peng and Deng assembled a team of engineers in Shanghai and began developing AgiBot's first humanoid robot at a rapid pace, leveraging Shanghai's established manufacturing supply chain and component ecosystem.[8]
The company was registered as AGIBOT Innovation (Shanghai) Technology Co., Ltd. (Chinese: 智元机器人, Zhiyuan Jiqiren, meaning "Intelligent Origin Robot"). The global brand name "AgiBot" was derived from combining "AGI" (artificial general intelligence) and "Bot" (robot), reflecting the founders' ambition to create robots powered by general-purpose intelligence rather than narrow task-specific programming.[9]
On August 18, 2023, AgiBot publicly unveiled the RAISE A1 at a launch event in Shanghai, just six months after the company's founding. During the presentation, a RAISE A1 unit walked onto the stage under its own power, demonstrating bipedal locomotion. Peng Zhihui presented the robot's technical architecture, including the PowerFlow joint motor, the SkillHand system, the EI-Brain AI framework, and the WorkGPT model. The event also showcased the robot performing industrial tasks such as bolt tightening and using interchangeable end-effector tools.[1][3]
The rapid development timeline attracted significant attention in China's technology sector. Peng described his design philosophy by stating: "We envision it as a capable worker rather than a mere human mimic," emphasizing practical industrial utility over humanlike appearance.[3] The robot's name, "RAISE" (远征, Yuanzheng, meaning "Expedition"), reflected the founders' view of the A1 as the beginning of a longer journey toward general-purpose embodied intelligence.
AgiBot secured substantial funding during and shortly after the RAISE A1's development period. In September 2023, the company closed a seed round of approximately $83.8 million (600 million yuan) led by HongShan (formerly Sequoia China), with participation from Hillhouse Capital and BYD. Additional investors included Warburg Pincus and Baidu. Within three months of the RAISE A1's launch, AgiBot's valuation reached 10 billion yuan (approximately $1.4 billion), making it one of the fastest Chinese startups to achieve "unicorn" status. By December 2023, the company had completed five funding rounds, accumulating over 600 million yuan in total financing.[10][11]
BYD's involvement as both an investor and a potential customer for AgiBot's robots proved particularly strategic, as the electric vehicle manufacturer would later deploy AgiBot robots in its welding workshops for logistics and quality control tasks.[12]
The RAISE A1 was always intended as a technology demonstrator rather than a mass-production product. After the August 2023 unveiling, AgiBot's engineering team used the A1 platform to refine its proprietary actuator, hand, sensor, and AI technologies. In January 2024, AgiBot established its manufacturing facility in Shanghai's Lin-gang Special Area, shifting focus toward developing commercially viable successors.[5]
In August 2024, AgiBot introduced five new robot models at a product launch event: the AgiBot A2 (standard bipedal humanoid), the A2 Max (heavy-duty industrial variant), the A2-W (wheeled dual-arm platform), and two compact models in the Lingxi series (X1 and X1-W). The A2 represented a direct evolution of the RAISE A1, incorporating lessons learned from the prototype while optimizing for commercial deployment, manufacturing scalability, and safety certification.[5][13]
Key differences between the RAISE A1 and its A2 successor included refined body proportions (the production A2 stands 169 cm and weighs approximately 69 kg), an upgraded sensor suite with 360-degree LiDAR and six high-definition cameras, improved walking speed (up to 3.3 m/s or 11.9 km/h versus the A1's 1.94 m/s), hot-swappable batteries for continuous operation, and a three-layer safety monitoring architecture that would enable the A2 to become the first full-size humanoid robot certified by China, the United States, and the European Union simultaneously in May 2025.[14][15]
The RAISE A1 was designed as a full-size bipedal humanoid with proportions approximating those of an adult human. AgiBot's engineering team prioritized industrial functionality, incorporating a modular architecture that allowed the robot to be configured for different task requirements.
| Specification | Value |
|---|---|
| Height | 175 cm (5 ft 9 in) |
| Width | 550 mm |
| Weight | 53 to 55 kg (117 to 121 lb) |
| Total degrees of freedom | 49+ |
| Hand DOF (per hand) | 12 active + 5 passive |
| Maximum walking speed | 7 km/h (1.94 m/s; 4.3 mph) |
| Total load-bearing capacity | 80 kg (176 lb) |
| Single arm payload | 5 kg (11 lb) |
| AI computing power | 200 TOPS |
| Battery life | ~2 hours |
Note: Some sources report the weight as 53 kg while others cite 55 kg. The discrepancy likely reflects different measurement conditions (with or without a full battery and end-effector tooling).[1][4][16]
The RAISE A1's locomotion system is built around AgiBot's proprietary PowerFlow joint motor, a quasi-direct-drive actuator designed to balance high torque output with lightweight construction and precise force feedback. Each PowerFlow unit weighs only 1.6 kg while delivering a peak torque exceeding 350 N-m. The actuator incorporates several key technologies:[4][17]
| Component | Description |
|---|---|
| Drive architecture | Quasi-direct drive with low cogging torque design |
| Gear reduction | High-transparency planetary reducer with less than 10:1 gear ratio |
| Position sensing | Coaxial dual encoders for precise joint angle measurement |
| Thermal management | Integrated liquid cooling circulation system |
| Motor control | Self-developed vector control driver |
The quasi-direct-drive approach offers advantages over traditional high-ratio gearbox designs commonly used in industrial robots. By using a low gear ratio (under 10:1), the PowerFlow actuator preserves the motor's natural backdrivability, meaning external forces applied to the robot's limbs are transmitted back through the motor with minimal friction. This property, often called "torque transparency," is critical for safe human-robot interaction and for enabling compliant, adaptive motion during manipulation tasks. The trade-off is that low-ratio drives require higher-torque motors to achieve the same output force, which AgiBot addressed through custom motor design and liquid cooling to sustain high currents without overheating.[4][17]
The PowerFlow technology established in the RAISE A1 was subsequently refined and deployed across AgiBot's entire product line, including the A2 series (which uses improved versions of the same actuator architecture) and the compact X1 and X2 platforms (which use a smaller "PowerFlow R" variant).[18]
One of the RAISE A1's most distinctive mechanical features is its reverse-jointed knee configuration, in which the knee bends in the opposite direction compared to a human leg (similar to the leg structure of ostriches or other digitigrade animals). This design choice expands the robot's range of motion in the sagittal plane, providing greater flexibility and a wider operational workspace for the lower limbs. The reverse-knee geometry also shifts the center of mass forward relative to the support polygon, which can improve dynamic stability during walking by keeping the robot's weight distributed favorably over the feet.[3][4]
Combined with the high-torque, backdrivable PowerFlow actuators, the reverse-knee design enabled the RAISE A1 to achieve stable bipedal walking at speeds up to 7 km/h and to maintain balance when subjected to external pushes or disturbances during demonstrations. This contrasted with the forward-knee (anthropomorphic) designs used by most competing humanoids at the time, such as the Unitree H1 and Fourier GR-1, both of which also launched in 2023.[3]
Notably, AgiBot shifted to a conventional forward-knee design for the successor A2 series, likely to improve compatibility with human-designed workspaces (such as stairways, vehicle interiors, and existing manufacturing fixtures) and to facilitate more natural-looking gait patterns for service-oriented applications.[14]
The RAISE A1's hands use AgiBot's SkillHand design, a five-fingered dexterous end-effector featuring 12 active degrees of freedom and 5 passive degrees of freedom per hand, with all drive mechanisms integrated inside the hand unit. The SkillHand was designed to support a wide range of manipulation tasks, from precision operations (fine alignment, tool use on small parts) to power grasps (bin handling, workstation operations).[4][19]
A defining feature of the SkillHand is its integrated visual fingertip sensing. Each fingertip contains miniature cameras capable of identifying the color, shape, and material of objects being grasped, while also providing simulated pressure sensing through visual deformation analysis. This approach reduces the cost and complexity of the hand by decreasing the precision requirements for mechanical components; instead of relying solely on expensive tactile sensors, the system uses vision-based methods to infer contact forces and object properties.[3][4]
AgiBot targeted a production cost of below 10,000 yuan (approximately $1,400 USD) per SkillHand unit, a significant reduction compared to competing dexterous robot hands that often cost tens of thousands of dollars.[3] The SkillHand technology was later succeeded by AgiBot's OmniHand series, which expanded the dexterous hand lineup for the A2 and G2 product families.[20]
The RAISE A1 was designed with a modular architecture allowing different end-effector tools and mobility bases to be attached depending on the application. In addition to the standard bipedal configuration, AgiBot indicated that wheeled and tracked base variants were possible. The arm-end interfaces supported interchangeable tools including screwdrivers, electric drills, and inspection instruments, enabling the robot to switch between tasks without hardware redesign.[3]
The RAISE A1 introduced AgiBot's first-generation AI software stack, centered on the Embodied Intelligent Brain (EI-Brain) architecture and the WorkGPT multimodal model. These systems represented AgiBot's approach to bridging the gap between high-level language understanding and low-level motor control in humanoid robots.
The EI-Brain is a four-tier computational architecture that divides robot intelligence into distinct functional levels, each responsible for a different abstraction layer of task execution:[4][21]
| Level | Component | Function |
|---|---|---|
| Mission level | Cloud Superbrain | High-level task planning, natural language understanding, long-horizon reasoning via cloud-based compute and WorkGPT |
| Skill level | Main Brain | Mid-level skill selection and sequencing, translating mission plans into executable skill primitives |
| Instruction level | Sub-brain | Low-level motion planning and trajectory generation for individual joint movements |
| Servo level | Brainstem | Real-time motor control, sensor fusion, and safety monitoring at the hardware level |
This hierarchical decomposition allows the robot to process a natural language instruction (e.g., "tighten the bolts on this panel") at the mission level, break it down into a sequence of skills (approach panel, grasp tool, locate bolt, apply torque) at the skill level, generate specific joint trajectories at the instruction level, and execute smooth, force-controlled motions at the servo level. The separation of concerns also enables different update cycles: the cloud-connected upper layers can be updated via software patches, while the real-time lower layers maintain deterministic control loops.[4][21]
WorkGPT is AgiBot's self-developed multimodal large language model that operates at the mission level of the EI-Brain architecture. It integrates large language model (LLM) and vision-language model (VLM) capabilities to enable the robot to understand user intent from natural language or visual cues, perceive the surrounding environment through camera feeds, and orchestrate multi-step task execution by selecting and sequencing appropriate skill primitives.[4][21]
AgiBot enhanced WorkGPT's perceptual and decision-making capabilities by training it on a proprietary robot task dataset, supplementing general-purpose language and vision training with domain-specific robotics data. This allowed the model to reason about spatial relationships, object affordances, and task-relevant physical constraints in ways that generic language models typically cannot.[21]
WorkGPT represented AgiBot's first foray into foundation model development for robotics. The company later developed more advanced models, including GO-1 (Genie Operator-1, launched March 2025), which introduced the Vision-Language-Latent-Action (ViLLA) framework, and GO-2 (launched 2026), which featured a dual-system architecture for high-frequency manipulation control.[22][23]
The RAISE A1 integrates a multi-modal sensor array for environmental perception:
| Sensor | Function |
|---|---|
| RGB-D camera | Color and depth imaging for 3D object recognition and scene understanding |
| LiDAR | 3D spatial mapping, obstacle detection, and environmental analysis |
| Microphone array | Audio input for voice commands and sound source localization |
| IMU (inertial measurement unit) | Body orientation and acceleration sensing for balance control |
| SkillHand fingertip cameras | Visual-based tactile sensing for grasp quality assessment |
The 200 TOPS onboard computing platform processes data from all sensors in real time, enabling the robot to maintain situational awareness while executing manipulation tasks. The robot also supports Wi-Fi (2.4 GHz and 5 GHz), Ethernet, USB, HDMI, and Bluetooth connectivity for remote monitoring, data upload, and cloud AI integration.[4][16]
The RAISE A1 was primarily designed for industrial applications in the automotive, 3C manufacturing, and logistics sectors. AgiBot demonstrated the robot performing several task categories during and after the August 2023 launch event:[1][3]
Beyond industrial deployment, Peng Zhihui articulated a long-term vision for the RAISE A1 platform (and its successors) as household assistants. He described three envisioned domestic roles: chef, nanny, and caregiver, with the robot capable of cooking, doing laundry, and providing elder care. However, Peng estimated that achieving reliable household deployment would require five to eight years of further development, placing this goal in the late 2020s timeframe.[3]
AgiBot set a long-term target of reducing the RAISE A1's hardware cost to below 200,000 yuan (approximately $30,000 USD), a price point intended to make humanoid robots economically viable for industrial customers. At the time of the 2023 launch, the actual estimated production cost was approximately 650,000 yuan ($90,000 USD) per unit, reflecting the early-stage nature of the supply chain and manufacturing processes. Approximately 80 to 90 percent of the robot's components were sourced from domestic Chinese suppliers.[3]
The RAISE A1 was not mass-produced. Instead, a limited number of prototype units were built for internal testing, demonstrations, and research purposes. AgiBot's mass production efforts focused on the successor A2 series, which entered production in late 2024 and scaled to over 10,000 cumulative units by March 2026.[12]
The RAISE A1 served as the direct predecessor to the AgiBot A2 series, and many of its core technologies carried forward into the production platform. The following table summarizes the key differences between the A1 prototype and the A2 production model:
| Feature | RAISE A1 (2023) | AgiBot A2 (2024) |
|---|---|---|
| Status | Prototype / demonstrator | Mass production |
| Height | 175 cm | 169 cm |
| Weight | 53 to 55 kg | ~69 kg |
| Degrees of freedom | 49+ | 40+ active |
| Maximum walking speed | 7 km/h (1.94 m/s) | 11.9 km/h (3.3 m/s) |
| Single arm payload | 5 kg | 15 kg |
| Knee design | Reverse-jointed (digitigrade) | Forward-jointed (anthropomorphic) |
| LiDAR | Single unit | 360-degree panoramic |
| Cameras | RGB-D | Six HD cameras (RGB-D, RGB, fisheye) |
| Battery system | Standard | Hot-swappable (700 Wh) |
| Safety certifications | None | CR (China), FCC (US), CE (EU) |
| AI model | WorkGPT | WorkGPT + ActionGPT + HIMUS 3D-SLAM |
| Price | ~$90,000 est. production cost | $100,000 to $130,000 |
| Units produced | Limited prototypes | Thousands (part of 10,000+ total AgiBot units) |
The A2's shift to a shorter, heavier frame with an anthropomorphic knee configuration reflected AgiBot's pivot from a research-oriented platform to a commercially deployable product. The A2 gained higher walking speeds through improved actuator design and gait optimization, while the addition of 360-degree LiDAR and multiple camera types provided more robust perception for autonomous navigation in unstructured environments. The hot-swappable battery system, absent from the A1, enabled continuous operation in factory settings.[14][15]
The A2 series also expanded into multiple specialized variants: the A2 Lite (cost-reduced for customer-facing roles), the A2 Ultra (enhanced perception), the A2 Max (heavy-duty industrial with 40 kg payload), and the A2-W (wheeled dual-arm platform for factory automation). This product diversification was made possible by the modular engineering foundation established during the RAISE A1's development.[13]
The RAISE A1 debuted during a period of intense activity in China's humanoid robotics sector. Several Chinese companies unveiled full-size humanoid robot prototypes in 2023, creating a competitive landscape that would define the industry's trajectory over the following years.
| Robot | Company | Height | Weight | DOF | Max Speed | Key Feature |
|---|---|---|---|---|---|---|
| RAISE A1 | AgiBot | 175 cm | 53 to 55 kg | 49+ | 7 km/h | PowerFlow actuators, WorkGPT AI |
| H1 | Unitree Robotics | 180 cm | 47 kg | 24 | 5.6 km/h (later 11.88 km/h) | Speed record, backflip capability |
| GR-1 | Fourier Intelligence | 165 cm | 55 kg | 40 | 5 km/h | Rehabilitation background, 100+ units delivered |
| Star1 | RobotEra | 171 cm | 52 kg | 44 | 6 km/h | Open-source components |
| CL-1 | LimX Dynamics | 130 cm | 45 kg | 22 | N/A | Compact design, strong locomotion |
Internationally, the RAISE A1's launch coincided with growing interest in humanoid robots from companies including Tesla (which had demonstrated its Optimus prototype in 2022 and continued development through 2023), Figure AI (which unveiled its Figure 01 in early 2023), and Agility Robotics (which expanded production of its Digit robot).[24]
The RAISE A1 distinguished itself from these competitors through several design choices. Its 49+ degrees of freedom was among the highest in the 2023 cohort, reflecting AgiBot's emphasis on dexterous manipulation over pure locomotion performance. The integrated WorkGPT language model was one of the first attempts to embed a multimodal large language model directly into a humanoid robot's control architecture for task-level reasoning. The PowerFlow actuator's 350 N-m peak torque in a 1.6 kg package represented a strong power-to-weight ratio for the quasi-direct-drive class. And the SkillHand's visual fingertip sensing offered a novel alternative to traditional force-torque and tactile sensor arrays.[1][4]
However, the RAISE A1's reverse-knee design was unconventional and potentially limiting for human-compatible workspaces, its single-arm payload of 5 kg was modest compared to industrial requirements, and as a prototype it lacked the safety certifications and manufacturing readiness that would be needed for commercial deployment. AgiBot addressed all of these limitations in the subsequent A2 series.[14]
The RAISE A1 holds historical significance as the product that launched AgiBot's trajectory from a six-month-old startup to one of the world's largest humanoid robot producers by volume. Several of its contributions proved foundational:
By March 2026, AgiBot had produced over 10,000 robots across its product line, capturing approximately 39% of the global humanoid robot market in 2025. The company was preparing for a Hong Kong IPO at a valuation of $5.1 billion to $6.4 billion. All of this traced back to the RAISE A1 prototype that walked onto a Shanghai stage in August 2023.[12][25]