RoboSense Releases Second-Generation Fully Solid-State Perception Platform, Aiming to Be the Data Entry Point for Physical AI 速腾聚创第二代全固态感知平台发布，要做物理AI数据入口｜最前线

TL;DR

• RoboSense released the E2, a second-generation fully solid-state perception platform powered by its self-developed "Peacock" SPAD-SoC chip, marking a shift from hardware supply to infrastructure services.
• The platform achieves three times the precision of its predecessor with a wider field of view, addressing the critical lack of high-fidelity 3D spatial data (distance, material, force) needed for Physical AI.
• Traditional 2D vision and simulation data are insufficient for real-world robotic tasks; E2 provides native high-precision 3D information to close the loop between perception, understanding, decision-making, and iteration.
• RoboSense is positioning itself as a key data entry point for Physical AI, collaborating with companies like Origen and Jianzhi Robotics to build perception infrastructure for home, industrial, and inspection scenarios.
• The strategy emphasizes full-stack self-research of chips to standardize data output specs at the source, ensuring consistent, high-quality spatial data for continuous model evolution.

Why It Matters

This development highlights a pivotal transition in the embodied AI industry where the bottleneck is shifting from algorithmic capability to the quality of physical world data. For AI practitioners and researchers, it underscores that high-precision 3D depth and spatial structure data are now recognized as essential "production factors" for training robust Physical AI models, moving beyond reliance on imperfect 2D images or simulated environments.

Technical Details

• Core Hardware: The E2 platform is built on the self-developed "Peacock" SPAD-SoC (Single-Photon Avalanche Diode System-on-Chip) and 2D VCSEL chips, enabling full solid-state architecture where signal transmission and processing occur at the chip level.
• Performance Metrics: Compared to the previous generation, the E2 series offers a significantly wider field of view and up to 3x higher precision, providing the granular spatial data required for complex operations like grasping and long-term inspection.
• Data Architecture: Unlike competitors who assemble discrete components, RoboSense uses a full-stack self-research approach to define detection precision and point cloud output specifications at the chip design stage, minimizing performance loss and ensuring standardized data assets.
• Application Scope: The technology is deployed in diverse robotics including lawn mowers, humanoid robots, quadrupeds, and drones, targeting complex real-world environments such as factories, parks, and homes.

Industry Insight

• Shift to Infrastructure: Sensor manufacturers must evolve from selling standalone hardware to providing integrated data infrastructure services that enable continuous model iteration. Companies that control the quality and standardization of raw spatial data will gain significant leverage in the Physical AI value chain.
• Data Quality over Quantity: The industry is recognizing that high-fidelity 3D data is a scarce resource. Investment and R&D should prioritize sensors and methodologies that capture rich physical interactions (force, texture, depth) rather than just visual appearance, as this data is critical for generalizing robot behaviors from simulation to reality.
• Standardization as a Moat: By defining data output standards at the silicon level, early movers can create ecosystem lock-in. Developers building Physical AI models will increasingly depend on specific sensor data formats, making compatibility and data consistency key competitive advantages.