Xiaomi Races Ahead of Tesla with Humanoid Robot Unveil; Baidu Secures China’s 1st Permits for Fared Driverless Robotaxi; SenseTime Launches Chinese Chess Robot
Weekly China AI News from Aug 8 to Aug 14
News of the Week
Xiaomi Unveils Humanoid Robot CyberOne
What’s new: Chinese smartphone giant Xiaomi surprisingly debuted the company’s first humanoid robot, up to a month before Tesla’s anticipated unveiling of Tesla Bot on its annual AI Day. Named CyberOne, the robot measures 177cm in height and 52kg by weight, with an arm span of 168cm. It is said to detect human emotion and create three-dimensional virtual reconstructions of the real world.
While CyberOne was still wobbling like a two-year-old baby in the demo — completely dwarfed in contrast with the agile Atlas humanoid robot from Boston Dynamics — the robot reflects “Xiaomi’s future technological ecosystem and a new breakthrough for the company”, said CEO Lei Jun. CyberOne’s AI and mechanical capabilities are all self-developed by Xiaomi Robotics Lab.
Why it matters: From autonomous vehicles to move your around, to humanoid robots to help pick up your laundry, a robotic future that promises to make your lives simpler is the next utopia tech companies are envisioning … and working on. Elon Musk said that he believes Tesla’s goal of a humanoid robot is the most important thing it’s working on. Amazon just acquired iRobot, the maker of Roomba, for $1.7 billion this week, adding vacuum cleaners to its portfolio of home robots. Not to mention Spot, a $75,000 robot dog, can already perform smart and highly resourceful tasks like patrolling parks or exploring an old mine.
Xiaomi’s autonomous driving: CEO Lei Jun also showcased Xiaomi’s progress in autonomous driving as part of its EV initiative, aiming to become an industry leader by 2024. Xiaomi plans to invest RMB 3.3 billion in the first R&D phase and has established an R&D team of more than 500 professionals. The company also unveiled an autonomous vehicle demo.
Baidu Secures China’s 1st Permits for Fared Driverless Robotaxi
What’s new: Chinese search engine and AI giant Baidu received permits from the governments of Chongqing and Wuhan, two mega southern Chinese cities, that allow them to deploy fully driverless robotaxis on open roads and charge passengers. The permits are widely recognized as the first-of-its-kind permission in China, a major regulatory approval.
Apollo Go, Baidu’s autonomous ride-hailing service, has begun offering fared robotaxis in geofenced areas in Wuhan from 9 am to 5 pm, and Chongqing from 9:30 am to 4:30 pm, with five Apollo 5th-gen robotaxis operating in each city. The areas of service cover 13 square kilometers in the Wuhan Economic & Technological Development Zone, and 30 square kilometers in Chongqing’s Yongchuan District.
Why it matters: The permits technically put Baidu at the same level as Waymo and Cruise, which have been offering paid fully driverless robotaxi in Phoenix and San Francisco respectively. Said Baidu executive Wei Dong, “We believe these permits are a key milestone on the path to the inflection point when the industry can finally roll out fully autonomous driving services at scale.”
More importantly, China’s Ministry of Transport issued a guideline the day after Baidu’s permits, saying China will “adapt to the development of autonomous driving technologies and encourage the regulated application of self-driving vehicles for transport services, while ensuring safety”, SCMP reported. A fully driverless robotaxi paired with a remote driver or safety operator can run on roads with light and controlled traffic conditions.
Other rivals: Baidu and Pony.ai also gained permission this July to run commercial driverless robotaxis with a required operator on the front passenger seat. Guangzhou-based WeRide.ai began a fared robotaxi service in Guangzhou — with a driver remaining behind the wheel — in 2019. AutoX, a Shenzhen-based startup, claimed they launched a fully driverless robotaxi service starting in January 2021, without specifying government permission.
SenseTime Introduces a Household Robot For Chinese Chess Playing
What’s new: Chinese AI giant SenseTime launched a physical robot that can play Chinese chess with your kids face-to-face, named SenseRobot. The robot is composed of a squircle robot head with an interactive display, a magnetic robotic arm that can move chess to different positions, a foundation, and a physical chess board.
SenseRobot offers different playing modes including AI Chinese chess learning, endgame challenge, Chinese chess breakthrough, and final challenge as well as multiple skill levels to choose from, starting with “basic” to explain Chinese chess to more than 100 endgame settings and 26 levels of matches, according to SenseTime.
Is it worth it: As the company’s first consumer tech product, SenseRobot is endorsed by Olympic gold medalist Guo Jingjing as chief product experience officer. Priced at RMB 1,999 — pretty expensive for a chess bot — SenseRobot doesn’t reflect technological advances that will blow your mind, but instead a fun piece of AI software-hardware integration.
GLM-130B: An Open Bilingual Pre-Trained Model
Researchers from Tsinghua University proposed GLM-130B, an open bilingual (English & Chinese) bidirectional dense model with 130 billion parameters, pre-trained using the General Language Model (GLM) algorithm. As of July 3rd, 2022, GLM-130B has been trained on over 400 billion text tokens (200B each for Chinese and English). The English model is better than GPT-3 175B (+5.0%), OPT-175B (+6.5%), and BLOOM-176B (+13.0%) on LAMBADA and slightly better than GPT-3 175B (+0.9%) on MMLU. The Chinese model is significantly better than ERNIE TITAN 3.0 260B on 7 zero-shot CLUE datasets (+24.26%) and 5 zero-shot FewCLUE datasets (+12.75%).
BungeeNeRF: Progressive Neural Radiance Field for
Extreme Multi-scale Scene Rendering
Researchers from the Chinese University of Hong Kong, Max Planck Institute for Informatics, University of Bath, and Shanghai Artificial Intelligence Laboratory have managed to bring Neural Radiance Field (NeRF), a neural network that created 3D views from a 2D input, to city-scale. They introduced CityNeRF, a progressive learning paradigm that grows the NeRF model and training set synchronously. The results showed the superiority of CityNeRF in modeling diverse city-scale scenes with drastically varying views, and its support for rendering views in different levels of detail.
AvatarGen: a 3D Generative Model for Animatable Human Avatars
Researchers from the National University of Singapore and ByteDance proposed AvatarGen, the first method that enables not only a non-rigid human generation with diverse appearances but also full control over poses and viewpoints, while only requiring 2D images for training. Benefiting from these designs, the method can generate animatable human avatars with high-quality appearance and geometry modeling, significantly outperforming previous 3D GANs.
Black Sesame Technologies, an AI digital imaging technology startup, has raised over $500 million in its Series C and C+ funding rounds. Founded in 2016, the Shanghai-based company develops AI chips that can power L3 and above-level autonomous driving.
Insilico Medicine, a clinical-stage end-to-end AI-driven drug discovery company, has announced the closing of its Series D2 round, bringing the total Series D financing to $95 million. Founded in 2014, the New York-Hong Kong-based company develops AI platforms that utilize AI technologies to discover novel targets and design novel molecular structures with desired properties.