Technology Author: Linyan Feng, PAN Zichun Jan 11, 2019 05:49 PM (GMT+8)

Horizon Robotics debuted its Matrix-based crowdsourcing High-Definition map data collection and positioning solution named NavNet, a LiDAR sensing solution and presented XForce platform designed for smart city and smart retail at CES 2019.

Horizon Robotics Booth at CES 2019. PHOTO: Credit to Horizon Robotics

Jan 9, 2019/EqualOcean/-Horizon Robotics (地平线机器人), a Chinese chipmaker debuted its Matrix-based crowdsourcing High-Definition map data collection and positioning solution named NavNet, a LiDAR sensing solution at CES 2019. Horizon also presented the XForce platform designed for smart city and smart retail, an edge AI computing platform based on Horizon BPU 2.0 processor architecture (Bernoulli architecture) FPGA (Intel A10, Field Programmable Gate Array).

Matrix the Platform

Horizon Robotics’s Matrix is selected as a CES 2019 Innovation Awards HONOREE in the Vehicle Intelligence and Self-Driving Technology product category. Matrix is a self-developed autonomous driving computing platform, based on the BPU (Brain Processing Unit) 2.0, a proprietary and high-performance AI processor architecture independently developed by Horizon to empower L3 and L4 autonomous driving. The product is now proving for large autonomous driving suppliers in mass production, the company claimed. Horizon emphasized the importance of multi-sensor fusion of the product. YU Kai (余凯), the founder and CEO of Horizon, indicated that the future autonomous driving belongs to multi-sensor fusion owing to single sensor can barely tell the obstruction a human or an animal, even Liar ( commonly used by Google, see more discussion about Lidar in this article), in an interview with EO Auto.

Trace back to April 2018, the company has published Matrix 1.0, a platform based on FPGA/BPU in the 2018 Beijing International Automotive Exhibition. Horizon claimed to change the architecture into Journey chip, and they were working on the newest Journey 2.0.

BPU 1.0 based on Gauss architecture, it has updated technology and transited to Bernoulli architecture for Matrix.

It seems a great improvement in performance, for Journey 1.0 is only applicable to L2 ADAS, however, since the company did not turn out a Journey 2.0 chip, We can hardly see it as a big step forward comparing to last year.

Autonomous Driving Solutions

The two solutions, NavNet and Lidar sensing solution, demonstrate Horizon’s strong abilities of real-time mapping and detection, respectively.

Based by the Matrix platform's edge computing capability, NavNet can conduct data collection, recognition, vectorization of 16 categories of map by deep learning and SLAM(simultaneous localization and mapping) technology, merely with the monocular camera visual perception application.

Horizon’s Lidar sensing solution provide real-time detection of  pedestrians and vehicles by CNN (Convolutional Neural Network) and skeleton point-cloud patch technology.

The company has cooperated with Tier 1 OEMs, Audi, Bosch, BYD, China Changan Automobile Group, Shanghai Automotive (SAIC).

Smart City and Retail Scenarios 

Apart from solutions for autonomous driving, Horizon presents the productized version of its XForce AI edged computing platform that is oriented to the scenarios of smart retail and smart city at the show, an equivalent to Matrix. The product was released in Oct 2018.

Facing different needs, Horizon provides edged computing FPGA XForce_P (Professional) and XForce_E (Express) respectively for the server market and personal computer market. Powered by the XForce edge AI platform and based on Sunrise 2.0 AI processor architecture, AIoT edge computing oriented multi-future urban scenarios, such as passenger flow statistics, automatic archiving, consumer action lines, scene thermograms, consumer behavior analysis, etc., can be used in various urban public areas such as transportation, security and so on, according to JQK News.

Horizon has always enjoyed the label of Embedded AI Processor and Processor+ Algorithm+ Cloud Platform and insisted on FPGA architecture, which is recognized as a main way for mass production in AI chip industry.

Even the company claimed the FPGA architecture products has a strong performance of high performance, low power consumption, easy integration and use of features compared to GPU (Graphics Processing Unit), in terms of power consumption, cost and performance, which still cannot beat NPU (Network Process Units).

Horizon was backed by Intel Capital, which led a USD 100 million investment for the company in 2017, changing technology route—which is FPGA-based chip designed by Intel—seems difficult for now, give so many products published and the deep correlation.