China’s robotics startup EngineAI is developing a control method that enables humanoid robots to walk with straight legs, mimicking natural human gait more closely.
Though still in the research phase, the technology is viewed as a crucial step toward achieving energy-efficient, stable, and agile movement in next-generation humanoid robots, according to a Chinese media outlet.
In February 2025, advanced control algorithms helped the firm’s compact offering, the PM01, become the first humanoid robot to perform a frontflip successfully.
Later on, the PM01 humanoid robot impressed viewers by flawlessly performing Kung Fu Hustle’s Axe Gang dance moves in a choreographed battle with professionals.
Smarter robotic walking
The Shenzhen-based robotics company EngineAI has applied for a patent on an innovative control system that allows humanoid robots to walk with straight legs, mimicking natural human motion, according to CGTN.
According to corporate databases Tianyancha and Qichacha, the system relies on an advanced AI framework incorporating human walking patterns, adversarial learning techniques, and real-world feedback to fine-tune the robot’s movements across diverse environments. Even though details of the new system are scarce, experts suggest the development marks a significant step toward creating more lifelike, stable, and energy-efficient robotic locomotion.
In April 2025, another Chinese humanoid, Adam, powered by a proprietary reinforcement learning algorithm, showcased human-like locomotion through extensive simulations and training data.
Conventional control strategies, including Model Predictive Control (MPC), employed in robots like Boston Dynamics’ Atlas and Spot, are effective in stable conditions but struggle in unpredictable environments. Through real-world interaction, robots can learn with the enhanced adaptability provided by deep reinforcement learning (RL). Nevertheless, its application in humanoids has been restricted due to expense and intricacy.
More from Innovation
See All
In response, PNDbotic created Adam, a humanoid robot that utilizes imitation learning and human gait data to achieve human-like motion. Adam is modular and motor-joint-driven. The modular design, biomimetic torso, and integrated full-stack system with an Intel i7-powered PND Robot Control Unit and real-time PDN network enable Adam’s advanced full-body motion control. This setup supports high-performance processing, neural network training, and dynamic simulations, making Adam well-suited for real-world service applications.
Agile robotic assistant
The PM01 is a small humanoid robot designed for researchers interested in investigating the combination of robotic and human-like abilities. The PM01, featuring a lightweight aluminum alloy exoskeleton, is 8 kg (17.6 lbs) lighter and 27 cm shorter than EngineAI’s flagship SE01, which measures 1.38 m (4.5 ft) and weighs 40 kg (88 lbs). The PM01, despite its smaller size, boasts remarkable features such as a 320-degree waist rotation and a bionic structure that replicates human movement with extraordinary fluidity.
The PM01 offers dynamic movement and the ability to carry out intricate tasks, thanks to its 24 degrees of freedom (DoF), of which 12 are in its joint motors. Its walking speed can attain 2 meters per second (4.5 mph), thanks to a quick-release battery that provides power for up to two hours on a single charge.
With its central touchscreen interface and sophisticated visual perception capabilities powered by an Intel RealSense depth camera, the PM01 offers high interactivity. It features an Intel N97 processor paired with an NVIDIA Jetson Orin CPU, allowing for real-time AI processing and high-performance computing. The PM01 is perfect for robotics research in settings that demand precision, agility, and smart interaction.
The robot’s business and education versions are priced at ¥88,000 (around $12,000), making them $3K cheaper than Unitree’s G1 humanoid.
