How NeoteAI’s Tactile Embodied AI Lets Robots ‘Feel’ the World – Near‑100 M CNY Angel Round

Tactile embodied intelligence is emerging as the missing link that lets robots move from merely seeing objects to truly feeling them during delicate operations such as plug insertion, cable handling, and fabric manipulation.

NeoteAI (Shanghai New Intelligence Embodied Technology Co., Ltd.) announced a close‑to‑100 million‑yuan angel round led by Shanghai Science & Technology Innovation Group and Fudan’s innovation fund, with additional backing from municipal and provincial programs that support its core tactile sensor research.

The company’s core team originates from the Fudan Trusted Embodied Intelligence Institute, including CEO Zhao Shihao (PhD, former researcher at Microsoft Research and Alibaba Tongyi Lab) and Chief Scientist Wu Zuxuan (former Meta engineer), bringing deep academic and industry expertise.

NeoteAI’s breakthrough is a self‑developed visual‑tactile sensor designed for grippers and dexterous fingers. The sensor captures force, slip, deformation, and boundary information during contact and converts these physical interactions into high‑density visual representations. By using monochrome illumination, particle‑based encoding, and model‑based decoupling, the design reduces lighting and camera constraints while keeping hardware costs low.

The accompanying data platform acts as a “fuel factory” for tactile models. It addresses the high‑cost, high‑complexity nature of tactile data collection by requiring end‑effectors equipped with sensing modules, synchronized force‑feedback loops, and comprehensive annotation pipelines. The platform has already amassed large‑scale datasets covering tasks such as precise plug‑in, assembly, and flexible material handling.

NeoteAI integrates this data into a VTLA (Vision‑Touch‑Language‑Action) model that fuses tactile modality deep within the model stack. Unlike traditional VLA frameworks that rely solely on vision and language, VTLA receives real‑time feedback on contact outcomes—whether a grasp is stable, a plug is fully inserted, or an object deforms—enabling the model to predict and adjust actions. The tactile world model reportedly lifts success rates above 90% in fine‑grained scenarios.

In reinforcement‑learning loops, tactile signals serve as a “golden signal” for policy refinement, allowing the system to correct unstable grasps or abnormal resistance on the fly, thereby driving failure rates toward zero for high‑difficulty tasks.

Industrial validation begins in factory environments where tasks have clear boundaries and measurable outcomes (success rate, cycle time, damage rate). NeoteAI has secured multiple proof‑of‑concept orders from sectors such as automotive, consumer electronics, and home textiles, demonstrating that tactile‑enhanced robots can handle operations that pure visual systems cannot.

With tactile perception positioned as an indispensable component for robots to operate reliably in the physical world, NeoteAI aims to open a new stage of “no blind spots, ultra‑precise manipulation” for embodied AI.