AI Vision Chip-Based 4-Camera Autonomous Navigation and Real-Time Trajectory Planning Technology
Technology Introduction:
This technology is an outcome of the MOEA Technology Development Program "Autonomous Development of Key UAV Chips and Modules," jointly developed by AzureWave Technologies (lead company) and Taiwan Swarm Innovation Inc. (partner company). Taiwan Swarm Innovation is responsible for the vision-based navigation, obstacle avoidance and trajectory planning algorithms, as well as full-system UAV integration and flight validation. Running on a Taiwan-made AI vision module and combining a four-lens fisheye camera set with a ROS2 software architecture, the technology enables UAVs to localize, avoid obstacles, navigate and track targets autonomously in GNSS-denied indoor spaces and interference-prone environments.Key Technology 1: 360-Degree Vision-Based Obstacle Avoidance with Visual-Inertial Navigation
Four fisheye cameras are fused with AI image perception and a Visual-Inertial Navigation System (VINS) to build a 360-degree depth point cloud, achieving a measurement error below 5% within a 5-meter range. The system performs autonomous localization and obstacle avoidance in GNSS-denied or heavily interfered environments, and is suited to indoor inspection, industrial facilities, bridge inspection and complex outdoor operations.
Key Technology 2: Real-Time Trajectory Planning and Autonomous Flight
Path-planning computation completes in under 100 milliseconds with an obstacle-avoidance success rate above 95%. In indoor flight tests, positioning error remains below 5% of the flight distance; the UAV traverses a 2 m x 2 m passage, avoids obstacles larger than 5 cm in diameter, and performs precision autonomous landing on a charging pad.
Key Technology 3: Built on a Taiwan-Made AI Vision Module Platform
The technology runs on the Taiwan-made AI vision module developed under this project, which delivers AI computing performance on the order of one hundred TOPS together with 8K high-resolution image processing, and provides a Reference Design Kit (RDK), SDK and ROS2 software framework that substantially lower the barrier to AI development for UAVs and robots.
Industrial Applications:
This technology is widely applicable to unmanned aerial vehicles (UAVs), robotics, intelligent inspection, smart logistics, precision agriculture, and public safety applications. By integrating AI-powered computer vision, autonomous navigation, and obstacle avoidance technologies, it enables autonomous inspection, monitoring, and target recognition in complex indoor and outdoor environments. The solution significantly reduces operational risks and labor costs while improving operational efficiency, safety, and mission reliability.Furthermore, the platform provides a comprehensive hardware and software development ecosystem, including a complete AI development framework, enabling domestic manufacturers and system integrators to rapidly develop and commercialize intelligent autonomous systems. This accelerates innovation and promotes the growth of the UAV and intelligent autonomous systems industries.
Taiwan Swarm Innovation Inc.
Founded in 2019, Taiwan Swarm Innovation Inc. is headquartered in the Shalun Green Energy Technology Demonstration Site in Tainan. With its core technology of self-developed UAV hardware and software, Taiwan Swarm Innovation Inc. is committed to providing customized UAVs, drone show technology development and demonstrations, UAV technology education and training, and diverse professional application services (such as inspection, monitoring, and aerial photography). The company's development is structured around four major areas: R&D, demonstrations, education, and services. It provides comprehensive and professional UAV system integration and services (UAS Turnkey Solution Provider), striving to become a long-term and trustworthy driving force for Taiwan's UAV industry.
Contact Information
Name：PR Lin
Tel：06-243-4700｜0910-651-529
Address：Rm. 501, Bldg. B, No. 360, Gaofa 2nd Rd., Guiren Dist., Tainan City 711, Taiwan (R.O.C.)
