Daily AI & Robotics Wrap: Humanoids Push Boundaries in Design and Intelligence
Xpeng Unveils ‘Most Human-Like’ Iron Robot with VLT AI Brain
Chinese electric vehicle and robotics company Xpeng showcased its next-generation ‘Iron’ humanoid robot at its 2025 AI Day event, emphasizing a design philosophy focused on maximum human-likeness. The robot’s demonstration, which featured a distinctly human-like, female form walking with a gentle, fluid gait, was so realistic that CEO He Xiaopeng noted colleagues had questioned whether there was a human inside the suit during rehearsals.
The new Iron robot incorporates a new, complex, multi-modal AI architecture that Xpeng calls its “brain,” aiming to move its robots from remote operation toward true autonomy. This architecture combines three large models: VLT (Vision-Language-Task), VLA, and VLM. The VLT model is described as the “core engine” for the robot’s autonomous actions and decision-making in real-world scenarios.
A key strategic pivot for Xpeng is the commitment to a human-like form, which the company believes is crucial for collecting “valuable data from human” environments like homes, offices, and shopping malls. The company had previously focused on four-legged designs but pivoted due to their difficulty in navigating complex, human-centric spaces and their lack of hands.
Hardware features designed to enhance its human-like capabilities include “dexterous hands” with 22 degrees of flexibility per hand and a human-like spine that allows the robot to bend over. The platform is also designed for customization, offering options for different body shapes and sexes, and includes a “full coverage soft skin” with touch sensors. Xpeng stated it is using an all-solid-state battery in the Iron robot, citing the stringent safety requirements for a household robot as the ideal application for the new technology.
1X Neo Robot Opens Preorders, Sparking Debate Over ‘Human-in-the-Loop’ Model
California-based company 1X has opened preorders for its humanoid robot, Neo, targeting the consumer home market. The $20,000 robot, which is expected to ship in 2026, is designed to perform various household chores like folding laundry, running a vacuum, and tidying shelves. The robot features a soft, human-like gait thanks to a tendon-driven motor system and is covered in soft materials to appear less intimidating than metallic prototypes from competing companies.
The company’s approach, however, relies on a “human-in-the-loop” operation, which 1X refers to as ‘Expert Mode.’ This strategy involves human teleoperators assisting the robot behind the scenes to help it learn safely and effectively. This model has drawn commentary from industry figures, including computer industry luminary John Carmack, who suggested that companies selling the dream of autonomous household robots should “embrace reality and sell ‘remote operated household help'” instead.
1X is positioning Neo as a “consumer-ready humanoid,” distinct from other robots like those from Tesla and Figure, which are currently focused more on factory environments. The company has also addressed privacy concerns, stating that Neo will only listen when addressed and its cameras will blur out human figures. Owners can also restrict the robot from entering or viewing specific areas of the home, and the robot will never be teleoperated without the owner’s approval.
Global Experts Convene in Shanghai to Address Humanoid Robotics Challenges
A forum on humanoid robot innovation and collaboration was held in Shanghai during the China International Import Expo (CIIE), highlighting China’s growing role in the global humanoid robotics sector. The event brought together industry leaders and experts to discuss both the recent breakthroughs and the key challenges facing the field.
The primary challenges discussed included data scarcity in the process of rapid skill iteration and the critical need to transition humanoid robots from technological demonstrations to practical, real-world applications. Experts noted that humanoid robots are becoming comprehensive platforms that combine perception, decision-making, and execution capabilities, increasingly integrating into human activities.
- China’s expanding robotics capabilities were evidenced by over 50 companies showcasing more than 100 products at the 2025 World Robot Conference, demonstrating applications from industrial analysis to precision handling.
- One expert emphasized that for robots to live alongside humans, they need to be human-sized and human-shaped, while also encouraging innovation beyond conventional humanoid designs.
- The founder and chairman of Unitree Robotics Co Ltd estimated that Chinese intelligent robotics companies would see average growth rates of 50-100 percent this year.
Caltech Researchers Develop X1: A Hybrid Humanoid That Walks, Flies, and Rolls
Researchers at the California Institute of Technology (Caltech) have developed a highly versatile hybrid robot, X1, which combines a Unitree G1 humanoid robot with Caltech’s M4 drone. This machine blurs the lines between a traditional humanoid, a drone, and a vehicle, giving it an unprecedented level of mobility.
The X1 system operates by having the humanoid act as a carrier platform, walking on two legs, climbing stairs, and manipulating objects with its hands. Once it reaches an open space, the humanoid leans forward to create a launchpad, allowing the M4 drone, which is attached to its back, to take off and fly independently. The M4 drone itself is capable of transforming upon landing by rotating its arms to use its circular propeller guards as wheels, allowing it to roll across the ground.
Integrating the two independent robotic systems proved to be a significant engineering challenge, which took the Caltech team three years to complete. The process involved modifying the Unitree G1 for autonomous navigation and extensively recalibrating the balance and control systems, as adding the drone shifted the humanoid’s center of gravity. The resulting X1 can travel across land and through the air, giving it a unique versatility in robotics.
Founding Split at 1X Leads to New Rival Firm Focused on ‘Physical Intelligence’
A significant strategic and personnel development occurred in the humanoid sector as Dr. Phuong Nguyen, the co-founder and “principal inventor” of 1X, split from the company to launch a rival robotics firm, Physical Robotics AS. This move represents a divergence in strategic approach for developing advanced AI for robotics.
While 1X is pursuing a consumer-focused “embodied learning” strategy with its Neo robot, Nguyen’s new venture is betting on what he terms “Physical Intelligence” (PI). This new approach will focus on gathering data from force-sensitive industrial manipulators rather than the general consumer environment. The split is framed not as a disagreement over the importance of AI, but rather a difference in the foundational method for building that intelligence.
Iowa State Researchers Prioritize Safety and ‘Physical Intelligence’ in Robotics Research
Researchers at Iowa State University are focusing on fundamental science to advance the “physical intelligence” and capabilities of humanoid robots while also prioritizing the development of safety standards for human-robot interaction. The work addresses the fact that physical intelligence—the ability to adapt to unpredictable environments and master complex motor skills—is notoriously difficult to replicate in machines, despite significant advances in AI.
Roboticist and Assistant Professor Bowen Weng noted that for humans, skills like walking and grasping become second nature, freeing up the mind for other tasks, but for robots, this physical intelligence is not automatic. The research aims to improve the efficacy and safety of robots through human-led studies and new algorithms.
“The bottom line is, you have to be able to trust it,” Weng said, “and the path to proving the trustworthiness of humanoid robots is through human-led research.”
Researchers are also addressing major hurdles to widespread adoption, including high development costs, a lack of standardization, and social and ethical challenges. The work emphasizes that humanoid robots still rely on humans for design, training, supervision, and ethical guidance, suggesting that new career paths in AI oversight and maintenance will emerge as human-AI collaboration becomes more common.