Horizon Matrix 2 platform for autonomous driving unveiled at CES 2020

LAS VEGAS — Horizon Robotics this week announced at CES here Horizon Matrix 2, the latest version of its computing platform for autonomous driving. The system builds on its predecessors, which have been adopted by mobility as a service (MaaS) providers, OEMs and Tier 1 suppliers for autonomous vehicle fleets around the world.

Founded in 2015, Horizon Robotics claimed to be is a pioneer in artificial intelligence for smart mobility at the edge. With more than 600 patents, the China-based company said it has been developing high-performance, low-power, and cost-effective deep-learning computing. The goal is to advance transportation safety, enhance the driving experience, and empower the future generation of autonomous vehicles.

The startup added that its partners benefit from its development toolkit and sets of pre-trained model examples for in-cabin applications and automated driving.

Horizon Matrix 2 supports visual perception

The passively cooled Matrix 2 uses Horizon Robotics’ automotive-grade Horizon Journey 2 AI processor and supports operation under a wide range of environmental conditions, according to the company. The computing platform includes field-tested algorithms to deliver highly accurate vision perception for applications ranging from driver-assist systems to fully autonomous robotaxis, shuttles, trucks, and delivery fleets.

Each Horizon Matrix 2 can support up to four high-definition, real-time perception video streams with a 3-Matrix system-level solution of 12 cameras (eight narrow field-of-view and four fisheye) to allow 360-degree vision perception.

Horizon Matrix 2 can be tailored for multiple use cases and operating design domains (ODD). The bundled perception software supports up to 23 classes of pixel-level semantic segmentation; comprehensive 2D/3D object detection; and classification of multiple vehicles types, pedestrians, cyclists, traffic lights and traffic signs.

In addition, the computing platform can support challenging, less-common yet important scenarios such as construction zones, said Horizon Robotics.

Customers interested in leveraging Matrix’s performance efficiency while deploying own proprietary deep learning models can utilize Horizon OpenExplorer AI toolkit. Horizon OpenExplorer supports popular deep learning frameworks such as TensorFlow and MXNet as well as state-of-the-art operators, backbones, and models. A comprehensive library of design examples and documents are also included to help customers reduce development efforts.

“With our Horizon Matrix 2 platform, our customers and partners can significantly reduce the development costs and accelerate time to market for their autonomous driving fleet,” said Yufeng Zhang, general manager of the automotive product line and vice president of Horizon Robotics.

Strategic partners

Horizon Robotics raised $600 million last February. Its partners include Audi, Chang’An, and Bosch. The company said it is also working with HD mapping companies such as SK Telecom, the largest telecommunication service provider in Korea, and DeepMap, a Palo Alto, Calif.-based startup, to provide HD mapping technology for autonomous driving.

At CES 2020, Horizon Robotics and Faurecia, a globally leading Tier 1 supplier, announced a strategic partnership to co-develop multi-modal AI perception solutions and accelerate commercialization for next generation intelligent cockpit systems.

“We are excited about the growth of our ecosystem of partners and customers around the world and are deeply committed to helping our partners and customer succeed,” stated Yufeng Zhang. “Smart mobility stands at the core of our company mission towards a safer and better life, and we welcome and appreciate the trust of all our partners and customers worldwide.”