SenseTime Unveils “Wu Neng” Embodied Intelligence Platform, Enabling AI to Evolve Through Real-World Interaction

At the “AI+: Large Model Shapes the Future” WAIC 2025 Large Model Forum organized by SenseTime, the company officially launched the “Wu Neng” Embodied Intelligence Platform. Built on SenseTime’s range of core technologies – including the SenseFoundry visual perception system, SenseCore, and SenseNova foundation models – the WuNeng platform is powered by the KaiWu World Model and supported by robust edge and cloud capabilities. It equips robots and intelligent devices with advanced perception, visual navigation, and multimodal interaction, accelerating the evolution of smart terminals toward higher degrees of autonomy and intelligence.

Dr. Xu Li, Chairman of the Board and CEO of SenseTime, said, “The integration of world models and embodied AI marks the next phase of AI development – a shift from tools to agents. With the ‘Wu Neng’ platform, we hope to support embodied AI companies in realizing their vision of interacting with the real world.”

Three Core Engines for Embodied Intelligence: Perception, Navigation, and Interaction

Perception, navigation, and interaction are the three key drivers of embodied intelligence, and SenseTime has deep technical foundations in all of them. These capabilities now converge on the “Wu Neng” platform. “Wu Neng” can be applied across a wide range of terminals, including cars and robots, to enable real-world interaction across physical spaces.

Perception is the foundation for machines to explore the real world. With over a decade of expertise in visual AI through SenseFoundry, Wu Neng can empower a wide range of end devices, from quadruped robots to humanoid agents, with the ability to recognize and understand the physical environment. The platform is highly adaptable across scenarios and supports efficient on-device inference through edge chip integration, enabling faster and more responsive perception.

Whether from the perspective of a robotic dog or a humanoid robot, the platform enables the seamless perception of surroundings and natural segmentation of objects.

Navigation is the cornerstone for machine mobility. Drawing from the robust visual-based navigation system of SenseTime’s SenseAuto end-to-end assisted driving solution, “Wu Neng” extends this capability to humanoid robots, quadruped robots, and more, enabling them to perform accurate path planning and obstacle avoidance across diverse environments.

From a robotic dog moving along a garden path to an autonomous vehicle navigating an urban lane, the platform ensures precise movement planning while avoiding obstacles automatically.

Interaction bridges the digital and physical realms. Built on the rapidly iterating SenseNova foundation model, “Wu Neng” allows robots to communicate with the real world in more natural and expressive ways. Its key strengths include emotional nuances, long-term memory, contextual awareness, and high stability.

In a demo, a humanoid robot powered by Wu Neng dynamically presents an AI-generated presentation for the film The Litchi Road, conveying various tones such as being humorous or formal.

Building a 4D Physical World: Supplying High-Quality Training Data for Machine Evolution

Powered by deep multimodal understanding, the KaiWu World Model developed by SenseTime ensures spatial and temporal coherence in video generation, producing high-quality data that enhances the intelligence of embodied devices. For example, KaiWu can generate realistic driving simulations from the perspectives of seven cameras, based solely on natural language prompts, with accurate geometric alignment across all perspectives and adhering to the laws of physics.

It can also “modify” the real world: replacing, deleting or adding scene elements like vehicles to create diverse physical environments.

Switching a sedan to a cargo van: Original video

Switching a sedan to a cargo van: Sedan replaced

With these modifying and generation capabilities, KaiWu simulates interactions like steering, braking, and acceleration, creating realistic driving videos that feel like “Need for Speed” in real-world settings.

Compared to smart vehicles, robots operate in more diverse environments and can actively generate high-quality physical data through exploration. In today’s data-scarce landscape for embodied AI, building a 4D world that models people, objects, and spaces becomes key to breaking the limits of model generalization.

Looking ahead, SenseTime aims to empower embodied AI companies to continue advancing in perception, understanding, and generation, transforming the challenge of uncertainty in embodied interaction to a clear path for industrial advancement, and accelerating AI’s evolution into the next decade.