Alibaba Develops Autonomous Driving: Champions of Three Tasks

Alibaba appeared on the road scenario segmentation list of the world’s largest autonomous driving computer vision algorithm KITTI on January 14.

iDST, Alibaba‘s artificial intelligence team, ranked first in three tasks, UU_ROAD (country lanes), UMM_ROAD task (multiple lanes) and URBAN_ROAD (urban lanes).

Alibaba confirmed the news to, but denied that Alibaba would enter the autonomous driving business.

Alibaba said it values basic scientific research and autonomous driving technology is a part of that. Alibaba‘s autonomous driving work is led by machine vision scientist Ren Xiaofeng.

KITTI is a compulsory platform that tests autonomous driving. Tucson, Uniview, Samsung and NEC were on the list before.

KITTI is jointly established by the Karlsruhe Institute of Technology and Toyota American Institute of Technology. It is the largest internationally recognized data set that tests computing algorithms for automatic driving.

KITTI covers five major scenes, city, residential areas, roads, campus, and pedestrians. It adopts BEV (Bird’s Eye View), which is more suitable for automatic driving. KITTI contains many difficult roads without signs.

According to KITTI’s official website, this evaluation data set is mainly divided into scene, target detection and target tracking. A researcher told scene segmentation is crucial in autonomous driving. It is mainly applied in recognizing areas for autonomous driving, planning routes, building precise maps and offering auxiliary driving AR (augmented reality) navigation.

Most autonomous driving research institutions are able to segment large sections of urban roads, but it is difficult to identify roadsides, such as stones and parking spaces along the road. How to divide roadside more finely is the main direction of the road scene segmentation.

Alibaba said based on online sample exploration, its iDST team increased its online data modules and established mechanisms to integrate comprehensive and local features. Thus, it was able to solve the problems of road segmentation, such as inaccurate segmentation between road and vehicles, and troubles caused by shadows.

After half a year research, Alibaba ranked first in UU_ROAD, UMM_ROAD and URBAN_ROAD with segmentation accuracy of 96.06 percent, 97.7 percent and 96.76 percent respectively. It is also the first time that Alibaba has been on the KITTI.

In June 2017, reported Ren, the top Chinese scientist at Amazon, joined Alibaba‘s iDST team. Ren dominated the Amazon Go algorithm. He was praised to have “comprehensive and in-depth understanding of image classification, object recognition, tracking and detection in the field of computer vision.” also found Alibaba iDST performed well in several other projects at KITTI, including vehicle detection, pedestrian detection and pedestrian reidentification. These technologies have landed in Alibaba Cloud ET City, which is mainly used for urban traffic management.

Alibaba has increasingly strong scientific power, and it emphasizes core science research. At the Computing Conference held in Hangzhou in 2017, Alibaba established Alibaba DAMO Academy (the Academy for Discovery, Adventure, Momentum and Outlook). Alibaba plans to invest more than 100 billion yuan in core science and technology innovation during the next three years. Alibaba chairman Jack Ma said he hopes DAMO Academy will be Alibaba‘s legacy gift to the world.

This article originally appeared in The Paper and was translated by Pandaily.