Brain-inspired technology to create new low-power, low-latency artificial intelligence systems is expected to make autonomous driving smarter and safer

 　　Since the birth of automobiles, human beings have begun to explore and imagine smart cars. Artificial intelligence drivers, thinking machines, robot cars... There are many expressions for smart cars, and their essence is the process of machines realizing perception, decision-making and operation. More than 100 years later, toward the general direction of realizing autonomous driving, the automotive intelligent industry, software and hardware iterations are constantly accelerating, and new technology tracks and huge opportunities are being born.

　　Among them, brain-like technology that imitates the way the human brain works is considered to be an innovative path for automobile intelligence. By 2025, brain-like chips are expected to become the leading computing architecture for cutting-edge artificial intelligence deployments, and are expected to replace GPUs. Alan, an expert in artificial intelligence technology at Gartner, said: "The industry is looking for new paths to develop artificial intelligence systems, while reducing power consumption." In the past two years, news of brain-inspired technology entering the smart car track has continued to come.

Mercedes-Benz tests the water-like brain chip, mining energy efficiency ratio, real-time performance and ADAS supplementary technology

　　In order to make cars see, hear and understand more intelligently, a series of sensors for assisted driving such as vision and voice are used more widely. In order to make the comprehensive perception, calculation and operation of the car more sensitive and smooth, the computing power demand for domain controllers and chips is also increasing.

　　It is worth noting that running a super sensing + computing system also means that considerable power consumption per hour is required. With the increasing popularity of new energy vehicles, car manufacturers have become more focused on the relationship between intelligence and vehicle energy consumption and battery life. It is against this background that the brain-like technology marked by low latency and low power consumption attracts manufacturers to try, and Mercedes-Benz has become one of the pioneers.

　　Brain-inspired technology was born in the 1990s, inspired by biological brains. Scientists are concerned that insects use close to zero computing power, but they can adapt to the natural environment and perform flight, foraging, cognition, decision-making and consciousness, showing much more powerful performance than existing artificial intelligence. However, it was only in recent years that issues such as Moore's Law, energy consumption and heat generation have driven technical experts to seek higher-performance computing, less data, and better storage technology. Brain-inspired technology has received more and more attention. It transmits and processes information in the same way as the biological brain, while only consuming ultra-low energy consumption.

　　In recent years, IBM and Intel have successively developed neuromorphic research chips that mimic the human brain. The research and development of commercial brain-like chips for the application side is led by innovative technology companies such as SynSense, and firstly focuses on edge computing applications that do not rely on the cloud but have extremely high requirements for efficiency.

　　According to reports, in 2020, Mercedes-Benz joined the Intel neuromorphic research community "to study how brain-inspired chips can improve energy consumption, speed, and accuracy for automotive-related artificial intelligence use cases." In January 2022, Mercedes-Benz launched the VISIONEQXX concept car, equipped with a neuromorphic chip from Brainchip, and used it in the wake-up system - the wake-up speed is 5 to 10 times that of traditional voice control. According to reports, this "thinking like a human" concept car can travel more than 1,000 kilometers on a single charge. It is expected that between 2024 and 2025, the functions and designs of the next-generation products built by Mercedes-Benz's new-generation and mid-size car platforms, It will be built on the basis of this concept car.

　　In addition, Mercedes-Benz recently cooperated with Terranet, a Swedish developer of assisted driving systems. The collaboration revolves around the Voxelflow sensor technology developed by Terranet using event cameras. According to reports, this technology can detect obstacles in front of the car within 5ms, as a supplement to ADAS systems, enhance radar, lidar and camera systems, and improve solutions in directions such as "ghost probes" outside the car. Terranet says the technology will "redefine safety" and aims to protect lives inside and outside the car, and has already begun road testing.

BMW explores brain-like + intelligent cockpit, dynamic vision solution enhances safety domain high interaction

　　Before the arrival of fully autonomous driving, the car has ushered in the intelligence of human-computer interaction - based on more accurate gesture, face, attention, voice and other recognition, the intelligent cockpit will achieve a more sensitive multi-modal high interact. The ideal state is that without human intervention, the in-vehicle system can actively observe, understand, and predict the state of the person, and even when the person is not aware of their own needs, prompt and respond in time to ensure safer driving.

　　In addition to the traditional solution, in April 2022, SynSense, a brain-like chip developer founded in Switzerland, announced that it is exploring with BMW, focusing on a SoC that combines dynamic vision sensors and brain-like processors - Speck.

　　It is reported that this dynamic vision SoC will be mainly used for the exploration of smart cockpits. "In order to ensure human-machine co-driving, it is necessary to continuously monitor the driver, safety officer and the surrounding environment for a long time. The low-power brain-like chip can greatly extend the monitoring time and achieve true always-on." SynSense Technology Dai Lan, chief algorithm engineer, said in an interview.

　　According to reports, this pure-end dynamic vision solution integrates the DVS event camera and the brain-like processor in the brain-like chip for the first time, simulating the processing method of image information from the biological retina to the brain. It is a set of very small "eye-like" + "Brain-like" system without additional cameras. The scheme uses an asynchronous circuit design and uses event-driven features to break the traditional static visual limitation limited by frames, and only capture key information (events) in the picture, which can reduce data by 10 times to 1000 times. Power consumption, and complete real-time visual interaction tasks on the terminal: such as gesture, gesture, and intent detection.

　　Compared with traditional imaging methods, dynamic vision technology also brings a higher dynamic range, can cope with recognition under harsher lighting conditions, does not connect to the cloud, does not capture biometric images, and can better protect privacy.

　　In 2021, another neuromorphic vision solution company, Prophess, will cooperate with DTS, a subsidiary of Xperi, to develop DMS brain-like solutions with event-driven technology. It is understood that Xper, Seeing Machines and Smart Eye are the same as DMS system (driver status monitoring) development companies. Likewise, this scheme only reports optical flow changes, i.e. the most critical dynamic information, such as human facial expressions. In this regard, the chief technology officer of Xperi said: "Real-time monitoring of difficult and perceived facial and eye movements can save lives."

　　According to the World Health Organization report, 80% of traffic accidents are caused by fatigue, while distracted driving is caused by 25% of traffic accidents are fatal. With the introduction of DMS supervision in Europe, the most critical monitoring functions such as driver status, head posture, eyelid opening, and line of sight direction correspond to the advantages of brain-like technology event-driven, real-time monitoring, and low power consumption. No matter what kind of driving style the car ushered in, reducing the shortcomings of human operation and reducing the danger of driving will be the eternal core of driving.

More companies focus on brain-like + autonomous driving

　　It can be seen that automobile giants such as Mercedes-Benz and BMW and emerging companies such as SynSense, Proofs, and Brainchip have successively launched actions in the field of brain-like. In addition, more companies have also launched related tests. It is reported that Ford and Vale o have started to try brain-like chip applications in the direction of ADAS. Shishi Technology has also announced that it has jointly released a synthetic aperture imaging technology based on brain-like event streams and image frames with Wuhan University at CVPR, which can shoot occluded scenes from multiple perspectives and achieve high-quality target imaging, aiming at robots and autonomous driving. and other fields. With its strengths in power consumption and real-time performance, the attempt of brain-inspired technology in the direction of high-speed obstacle avoidance is also expected to form a differentiated and complementary development with solutions such as cameras, millimeter waves and lidars. As autonomous driving technology gradually matures, brain-like technology will inevitably usher in more companies trying to tap its application value, and its future performance is worthy of attention.