Gasgoo Munich-The 2026 World Artificial Intelligence Conference (WAIC) concluded recently. RoboSense, a global LiDAR leader, announced a strategic upgrade. The theme was "Robot Eyes, Building the Data Entry for Physical AI." The company is shifting from a hardware supplier to a Physical AI infrastructure provider.
RoboSense launched its second-generation solid-state perception platform, the E2. It also secured strategic partnerships. Partners include Zhizai Wujie, Jianzhi Robotics, Origen, and Guanglun Intelligence.
These moves outline RoboSense's entry into the upstream Physical AI sector. One involves hardware, the other software.
One Mass-Production Product, Four Ecosystem Deals
RoboSense debuted the E2, a second-generation solid-state digital LiDAR. It uses the in-house "Peacock" SPAD-SoC chip. The E2 is a flexible product family. It adapts to multiple scenarios rather than being a single model.
Image Source: RoboSense
The E2 leverages solid-state scanning architecture. It integrates emission, reception, and processing. This achieves a comprehensive leap in precision, field of view, size, and reliability.
A demonstration showed the E2 outputs sharper edges than the E1. It captures depth and contours with improved precision. This delivers near-image-level 3D perception.
The E2 family extends beyond automotive LiDAR. It equips robots, drones, and lawnmowers. It supports avoidance and mapping. It also serves as a core data collection unit. This provides standardized 3D input for Physical AI infrastructure.
The E2 acts as a front-end data entry. It connects the physical world with AI models.
Image Source: RoboSense
The show featured industrial ecosystem achievements. Top players like Zhiyuan, Unitree, Pudu, Youibot, and Geek+ showcased equipped machines. The Zhiyuan Lingxi X2 demonstrated interaction. Pudu's robots presented collaborative avoidance. Youibot and Geek+ displayed coordination capabilities.
This lineup across services and logistics demonstrates RoboSense's ability to scale hardware.
RoboSense cemented partnerships with Zhizai Wujie, Jianzhi Robotics, Origen, and Guanglun Intelligence. The cooperation spans the industry chain. It covers perception, collection, and training.
These agreements aim to build a network for 3D spatial data. RoboSense is integrating with data providers and model firms. This bridges the gap from hardware collection to training. It pushes embodied intelligence toward commercial use.
The partnership with Origen signals expansion into overseas markets.
Entering the Data Track: RoboSense's Underlying Logic
RoboSense's shift addresses a core pain point: the "data quality dilemma". Embodied intelligence currently faces severe data quality issues in model training.
Xie Tiandi, Marketing Director, explained past AI models used 2D video. Algorithms inferred depth to simulate 3D structures. This "guesswork" has flaws. Edges are blurry and relationships distorted. Errors include people appearing in front of vehicles.
World models trained on such data struggle to understand physical environments. Robots cannot make autonomous decisions in complex scenarios.
RoboSense believes the key is upgrading sensor solutions. LiDAR uses the dToF principle. It outputs precise 3D point cloud data. This accuracy exceeds visual estimation. It solves data distortion at the source.
However, LiDAR-equipped terminals are limited. This causes a shortage of 3D data. Traditional LiDAR hardware is rigid. It struggles to adapt to customized data collection equipment.
Image Source: RoboSense
RoboSense's in-house chip offers a solution. The "Peacock" SPAD-SoC integrates transceiving and processing. It is compact and flexible. It can be embedded into data terminals. The chip outputs standardized 3D data. This reduces cleaning and labeling workloads. It boosts pipeline efficiency.
Xie Tiandi outlined a "dual-loop" business layout. The first loop is hardware supply. It provides LiDAR to automakers. This is RoboSense's current foundation. It drives mass production and cost advantages.
The second loop is data infrastructure. RoboSense provides hardware to AI model companies. It moves into the upstream AI training link. This unlocks a second growth curve.
The two loops are mutually reinforcing. Hardware shipments cut costs and accelerate chip iteration. This proliferates data collection devices. The data business enhances RoboSense's role. It transforms the company into an infrastructure provider. It connects "Real World — Perception Hardware — Data — Model — Robot".
Entering the data sector is a natural extension. Competition is shifting from model capabilities to data supply. Upstream vendors with 3D perception technology can redefine industrial value.
Long-term industrial dividends from data entry hold greater potential than one-off hardware sales.
Conclusion
RoboSense's commitment to perceiving the physical world remains constant. Its business boundaries have shifted from LiDAR to data entry.
Challenges remain. Solid-state solution costs must drop. Data service models need polishing. Building a 3D data supply system will take time.
High-quality spatial data will become critical. Players securing an early position at the data entry will gain an advantage. RoboSense's move explores its second growth curve. It reflects the construction of China's Physical AI infrastructure.
