A Shanghai-based robotics company has unveiled a compact humanoid robot that marks the firm’s expansion beyond industrial collaborative robots into the broader field of intelligent robotics.

JAKA Robotics’ Pi is a compact humanoid robot that stands 1.22 meters tall and weighs 92 pounds (42 kilograms).

According to Jaka, Pi is designed for versatile real-world applications. The platform combines mobility, advanced perception capabilities, and human-like interaction to operate in a variety of environments.

Recently, China has introduced humanoid robots into its postal logistics network, using automated parcel sorters in Guangzhou to boost warehouse efficiency.

Compact humanoid debuts

JAKA Pi is a compact humanoid robot designed to showcase the company’s latest advances in embodied intelligence, motion control, and AI-powered interaction.

Measuring 1220 × 420 × 220 millimeters and weighing just 92 pounds (42 kilograms), the platform is among the most compact humanoids in its category.

The JAKA Pi features 27 degrees of freedom and newly developed integrated joint modules that are 15 to 27 percent smaller than previous generations, enabling a more compact and lightweight design. Its knee joints deliver up to 120Nm of torque for stable locomotion, while each arm supports payloads of up to 3 kilograms for object handling and manipulation tasks.

At the core of the robot is JAKA’s fusion brain architecture, built on Intel’s heterogeneous computing platform. The system separates high-level intelligence from low-level motion control. The “cerebrum” processes AI reasoning, vision perception, large language models, and application logic, while the “cerebellum” handles real-time movement through an EtherCAT-based control network operating with millisecond-level latency.

According to the firm, the dual-domain architecture enables the robot to interpret spoken instructions, understand its environment, generate action plans, and execute physical tasks with coordinated motion. By combining advanced AI with deterministic control systems, JAKA Pi serves as a versatile research and development platform for embodied intelligence, human-robot interaction, and next-generation robotics applications.

Beyond industrial automation

JAKA Robotics is a Shanghai-based robotics company founded in 2015 and best known for its collaborative robots (cobots) and emerging embodied AI platforms. Over the past decade, the company has evolved from an industrial automation specialist into a developer of intelligent robotic systems that combine advanced perception, force control, machine vision, and artificial intelligence.

Its core product lineup includes the JAKA Zu Series (Zu3, Zu5, Zu7, Zu12, Zu18, Zu20, Zu30), designed for general industrial automation tasks such as assembly, machine tending, palletizing, and packaging. The JAKA Pro Series (Pro5, Pro12, Pro16) is built for harsh industrial environments, featuring IP68-rated protection against dust, oil, and water.

For applications requiring precise force interaction, JAKA offers the S Series (S5 and S12), which integrates high-accuracy force sensing and advanced force-control capabilities. The AL and A Series combine robotic manipulation with machine vision, enabling perception-driven automation and easier deployment in dynamic production environments.

The company also produces compact robots such as the MiniCobo and Mini 2, aimed at education, research, hospitality, and small-scale automation. Supporting technologies include the JAKA Lens 2D and JAKA Lens VPS vision systems, six-axis force sensors, RoboHub control platforms, and low-code programming tools.

In embodied intelligence, JAKA has introduced the K-Series humanoid platforms, including the K1, K1L, and K1W, as well as the recently unveiled JAKA Pi humanoid robot. These systems integrate large language models, machine vision, force control, and real-time motion planning, positioning JAKA as a developer of next-generation AI-powered robots capable of operating beyond traditional industrial settings.

💾

Researchers at the University of Colorado Boulder (CU Boulder) are using digital twin technology and virtual reality (VR) to develop robots capable of supporting future lunar exploration missions.

The project centers on Armstrong, a small three-wheeled robot that can be remotely operated through an immersive VR interface, allowing users to perform tasks such as picking up and moving objects.

While still confined to laboratory testing, the system is designed to help engineers study how fleets of robots could one day work alongside astronauts on the Moon, assisting with construction, scientific research, and the development of future lunar habitats.

“Our efforts at CU Boulder are intended to make lunar robots more efficient and recoverable from errors, so precious astronaut time on the lunar surface will be better utilized,” said the team in a statement.

Training lunar robots

Researchers are exploring how digital twins—highly realistic virtual reality simulations—can train operators to control robots in the Moon’s challenging environment. The technology enables realistic practice in low-gravity, crater-filled terrain without risking costly lunar hardware or mission-critical equipment.

At the center of CU Boulder’s effort is a compact three-wheeled robot equipped with a robotic arm and claw capable of manipulating objects. While the platform operates in a laboratory environment, it serves as a testbed for technologies that could eventually support large-scale lunar exploration and infrastructure development.

The project focuses on a major challenge facing future Moon missions: enabling astronauts and operators on Earth to effectively control robotic systems under harsh, unfamiliar lunar conditions. The Moon presents unique operational challenges, including low gravity, rugged terrain, deep craters, and permanently shadowed regions, which can complicate navigation and task execution.

To address these challenges, researchers developed a highly detailed digital twin of the robot and its surroundings. A digital twin is a virtual replica of a physical system that mirrors its behavior in real time. Using the Unity game engine, the team recreated the robot’s operating environment with high accuracy, including its movement characteristics and interactions with objects. The virtual model was calibrated to ensure that the robot behaved in the digital environment exactly as it did in the real world.

The digital twin was integrated with an immersive virtual reality interface, allowing operators to experience robot control from a first-person perspective through onboard cameras. This setup enables users to practice complex manipulation tasks in a risk-free environment before operating physical hardware.

Virtual exploration training

To evaluate the effectiveness of the technology, researchers conducted experiments in which participants used the robot to perform precision object-handling tasks.

Some operators are first trained in the virtual environment before transitioning to the physical robot. Results showed that users who practiced with the digital twin completed tasks significantly faster and reported lower stress levels compared to those who only used the real robot.

The findings suggest that digital twins can become valuable training tools for future lunar operations, reducing learning curves and improving mission efficiency. Such capabilities are particularly important for space missions where robotic systems may cost millions of dollars and where operational errors can have serious consequences.

Building on the initial success of the indoor digital twin, researchers are now creating more advanced virtual models of lunar vehicles operating on the Moon itself. These simulations aim to replicate challenging environmental factors, including uneven terrain, lighting conditions, and lunar dust behavior.

Modeling lunar dust remains one of the most difficult technical challenges. As rovers travel across the surface, dust can be kicked into the air, potentially obscuring cameras, degrading sensors, and affecting vehicle performance. Because real-world lunar dust data is limited, accurately simulating its movement remains a key area of ongoing research.

According to researchers, by allowing operators to train in realistic virtual environments before deploying physical hardware, the technology could play a crucial role in enabling safer, more efficient robotic operations during future lunar missions and the long-term establishment of human infrastructure on the Moon.

💾

Vietnamese technology company VinDynamics has unveiled its first humanoid robot, Dyno, marking the country’s entry into the global humanoid robotics arena.

The intelligent robot made its international debut at the ongoing ICRA 2026 in Vienna and Computex Taipei 2026.

Designed as a versatile assistant for modern environments, Dyno combines advanced AI with a sophisticated sensing system to operate in dynamic settings.

According to the firm, the humanoid is intended for security and surveillance tasks in urban and commercial spaces while also being developed as a household assistant.

Vietnam’s humanoid leap

Dyno is an intelligent humanoid robot built to combine advanced artificial intelligence, environmental perception, and dexterous manipulation in a single platform.

Designed for operation in complex real-world environments, Dyno integrates a sophisticated AI system with a highly responsive sensor suite that enables autonomous navigation, situational awareness, and human-robot interaction.

Though detailed specifications are not out, the humanoid is being developed for applications ranging from security and surveillance in urban spaces, campuses, and service complexes to domestic assistance tasks requiring precise object handling. Its flexible arm architecture and advanced manipulation system allow it to perform a variety of interactions in dynamic environments, according to a statement by the firm.

A major demonstration at the ongoing ICRA 2026 and Computex 2026 highlights Dyno’s capabilities as an autonomous robotic guide. During pilot deployments at Vinpearl Safari Phu Quoc, the robot operated in challenging outdoor conditions, utilizing multilingual speech capabilities, natural language interaction, and real-time environmental awareness to engage with visitors and respond to questions.

According to VinDynamics, the deployment demonstrated the platform’s ability to function reliably in unpredictable service environments while maintaining continuous interaction with people.

Building better humanoids

Alongside Dyno, VinDynamics is showcasing a range of technologies that form the building blocks of its humanoid robotics platform. These include a high-performance actuator joint, a human-like robotic hand, and a dedicated AI training dataset designed to help robots learn and perform more effectively in real-world environments.

One of the key components is the VDM 80 actuator joint, a compact but powerful motor that acts as the robot’s muscles. Weighing less than one kilogram, it delivers high torque and precise movement while operating on a standard 48V power supply. The actuator supports industrial communication standards such as CAN FD, RS485, and EtherCAT, making it easy to integrate into robotic and automation systems. It can reach speeds of up to 235 rpm and is designed for long-term reliability, with an expected operational life of more than 10,000 hours.

VinDynamics is also presenting a robotic hand built to closely replicate the movement of a human hand. Featuring 11 moving joints and six actively controlled degrees of freedom, the hand can perform smooth and precise motions while handling a variety of objects. Integrated force sensors help improve grip control and accuracy, allowing the hand to carry out delicate tasks with consistent performance.

According to VinDynamics, when combined with the company’s AI training platform, these technologies provide the foundation for future humanoid robots capable of operating safely and efficiently in real-world settings.

💾

Chinese robotics firm Pudu has unveiled a next-generation industrial semi-humanoid robot designed for manufacturing and factory environments.

The D7 robot combines industrial-grade hardware with embodied AI, powered by the company’s PuduFM 1.0 foundation model.

Unlike conventional industrial robots that follow fixed programs and predefined workflows, the PUDU D7 can understand tasks, learn from experience, and adapt its actions in real-world settings.

The Shenzhen-based company says the platform is designed to improve factory automation by enabling intelligent decision-making, continuous learning, and greater operational flexibility across a wide range of industrial applications.

In 2024, Pudu unveiled Pudu D7, its first semi-humanoid robot, combining an omnidirectional chassis, robotic arms, and a human-like upper body.

Adaptive warehouse worker

D7 is a semi-humanoid robot designed to evolve through real-world experience rather than relying solely on pre-programmed capabilities.

At the core of the system is an end-to-end data collection architecture that captures and processes operational data across a wide range of industrial workflows. Through low-latency transmission and synchronized data processing, the platform creates a continuous feedback loop that supports ongoing model training and performance optimization.

Powered by the PuduFM 1.0 AI foundation model, the robot learns from real-world operations, gradually improving its understanding of industrial procedures, spatial relationships, and manufacturing processes. Unlike traditional robots whose capabilities remain fixed after deployment, D7 continuously enhances its performance through every task it performs.

According to a statement by the firm, D7 is designed to improve efficiency in warehouses, factories, and logistics environments. Equipped with dual robotic arms, the robot can autonomously perform tasks such as material handling, shelf picking, inventory replenishment, and internal transportation. It supports payloads of up to 30 pounds (14 kilograms) and can operate at heights of up to 6.5 feet (2 meters), allowing it to interact effectively with high-rack storage systems.

Its force-control technology enables careful handling of both heavy and delicate items, reducing the risk of damage during transport. The robot can also push carts and move materials across facilities. By integrating navigation, approach planning, grasping, and manipulation into a unified action framework, PUDU D7 eliminates the traditional separation between movement and operation, enabling more efficient execution of complex, multi-step tasks.

Smarter industrial intelligence

For applications requiring fine control, such as assembly, dispensing, and precision handling, D7 incorporates advanced tactile sensors that provide real-time force and pressure feedback. The system delivers millimeter-level force-control accuracy, automatically adjusting applied force according to task requirements and material characteristics. This enables stable, consistent operation while minimizing the risk of component damage.

To operate safely in dynamic industrial settings, the robot is equipped with dual front-and-rear LiDAR systems and a comprehensive perception suite. These technologies allow it to continuously monitor its surroundings, detect obstacles, adapt to route changes, and navigate safely through busy factory environments with minimal infrastructure modifications.

PUDU D7 supports fully autonomous battery swapping, allowing it to independently replace and recharge batteries without human intervention. This capability minimizes downtime and enables continuous 24/7 operation, improving equipment utilization and productivity.

According to the firm, looking ahead, PuduAgent, its embodied AI agent platform, extends this vision by enabling robots to interpret user goals, break down complex workflows, coordinate navigation and manipulation tasks, and collaborate with multiple robots across different locations. Together, these technologies represent a shift from task-based automation toward autonomous systems capable of decision-making, long-term planning, and intelligent collaboration in industrial environments

💾