Car stuff (public number: chedongxi)
Author | Xiaofang
Edit | Juice
In the second half of today's intelligent vehicle forum in the second half of the intelligent vehicle forum, technological innovation accelerates the transformation of intelligent vehicles, Shen Shaojie, head of DJI Automotive, Wang Xiaogang, co-founder of SenseTime, and Gu Weihao, co-founder and CEO of Milli Zhixing, and other industry bigwigs participated in the forum and made speeches.
Based on cost control, DJI Automotive released a new generation of DJI Automotive intelligent driving solution, which can use a computing power as low as 32TOPS to realize the urban memory driving function that does not rely on high-definition maps, and at the same time, according to the cost calculation given by DJI Automotive, the cost range of its L2+ intelligent driving system is 5000~15000 yuan.
SenseTime, through the comparison of existing artificial intelligence and general artificial intelligence, pointed out that general artificial intelligence can use natural language to describe tasks, so as to give logical reasoning, which will be more flexible and intelligent than existing artificial intelligence.
Finally, the concept of heavy perception and light map is very popular this year, saying that it is now in the data-driven 3.0 era, and at the same time, it will also consider the price level and adopt an integrated solution of mooring.
The following is the dry content of DJI Automotive SenseTime and Milli Mobility in today's Smart Car Forum.
First, as low as 32TOPS computing power to realize urban memory driving
Shen Shaojie, head of DJI Automotive, shared on "breaking through the boundary between performance and cost and providing practical intelligent driving solutions for the Chinese market", in general, it is to squeeze the computing power of hardware to the limit, instinctively reject the stacking of sensors, weaken the dependence on external data, and reduce the assumption of the real environment to achieve a higher level of intelligent driving.
▲ Shen Shaojie, the person in charge of DJI Automotive
At the same time, DJI Automotive gives the total cost of L2+ intelligent driving system to the total price range, about 3%~5%, through rough calculation, L2+ intelligent driving system cost range of 5000~15000 yuan.
▲The cost range of L2+ intelligent driving system given by DJI
In this regard, Shen Shaojie also said that DJI can not only match high-resolution visual sensors with a system that integrates transportation and parking, but also ensure that the software has a high enough profit, and even when the budget is sufficient enough, it can also be equipped with lidar.
Therefore, in order to achieve cost compression and achieve better intelligent driving solutions, DJI's ideas are generally divided into four points: multi-scenario hardware multiplexing, extreme compression of hardware computing power, weak dependence, less assumptions, and in-depth industrial cooperation.
Multi-scenario hardware multiplexing and extreme compression hardware computing power can be summarized in one sentence, that is, improve the limit utilization rate of sensors, so that hardware costs can be reduced, and more scenarios can be applied with the least amount of hardware.
Weak dependence and low assumption do not rely on high-definition maps to realize L2+ intelligent driving functions including urban memory driving and urban pilot driving.
Finally, for in-depth industrial cooperation, it can adapt to different chip platforms, continuously expand the breadth and depth of cooperation with the upstream and downstream of the intelligent vehicle industry, and do a good job in the role of intelligent driving Tier1.
It is worth noting that based on the current cost constraints on intelligent driving systems, Shen Shaojie announced DJI's new generation of intelligent driving solutions, which can realize memory driving functions including urban areas with a computing power as low as 32TOPS and a pure visual configuration of 7V/9V.
▲Shen Shaojie introduced the 7V sensor configuration scheme
Among them, the 7V sensor configuration scheme is the basic configuration, including a pair of DJI's unique forward-looking inertial stereo stereo cameras, a rear-view monocular camera, and four surround-view fisheye cameras, which can realize all functions except urban pilot driving without relying on high-definition maps. At the same time, after using high-definition maps, the 7V configuration scheme can also realize urban piloting.
▲9V sensor configuration scheme (Source: DJI Automotive Official)
The 9V sensor configuration scheme with a computing power of 80TOPS improves the ability of dynamic vehicle detection on the side by adding two side-looking monocular cameras, which can realize urban pilot driving without relying on high-definition maps.
Finally, DJI's 7V/9V sensor configuration solution is actively promoting mass production with partner car companies, and DJI Automotive also revealed at this forum that it is preparing for a higher level of intelligent driving.
Second, general artificial intelligence Autonomous driving 3.0 era
Wang Xiaogang, co-founder, chief scientist and president of SenseTime, shared the challenges brought by general artificial intelligence and large models to smart cars.
▲Wang Xiaogang, co-founder of SenseTime
In terms of what is general artificial intelligence, Wang Xiaogang compared it with existing artificial intelligence systems.
Existing artificial intelligence has a preset control strategy, and for situations other than preset, it is necessary to re-collect a large amount of data and redesign the system. General artificial intelligence, on the other hand, generates multimodal data by entering prompt words, and more importantly, it can generate descriptions of tasks in natural language, give a series of logical reasoning, and finally give conclusions.
Finally, Wang Xiaogang said that SenseTime will release a series of multimodal large models on the Technology Product Day this month, including language large models and perception large models.
Gu Weihao, co-founder and CEO of Milli Zhi Heng, shared the practice of big computing power, big model and big data in the era of autonomous driving 3.0, showing that through heavy perception, large computing power, and no reliance on high-precision maps, the automatic driving of Milli will go faster.
▲ Gu Weihao, co-founder and CEO of Momo Zhixing
Gu Weihao said that from the perspective of industry technology trends, and from the perspective of comprehensive cost performance, there will be a more profound solution for driving and parking, combining the driving controller and parking controller into one, and the sensor is used comprehensively when driving and parking.
At the same time, the technology trend is divided into three eras, one is the hardware-driven 1.0 era, the body is covered with consumer-grade computing hardware with large computing power; the second is the software-driven 2.0 era, the highest height of software engineers determines the shortcomings of the product; the third is the data-driven 3.0 era, around the data layout application side.
▲Three eras of autonomous driving
Now is the data-driven 3.0 era, which needs to rely on the flow of data to produce breakthroughs in products and form a data-driven technology form.
Finally, Gu Weihao said that the self-driving generative large model DriveGPT released at the beginning of this year will be detailed at the AI Day on April 11, and Gu Weihao also said that Mumo has made a "6P model", that is, the solution from code to the whole can be opened to industry partners.
Conclusion: Technological innovation accelerates the transformation of intelligent vehicles
Whether it is the lower-cost and more practical intelligent driving solution brought by DJI, or the general artificial intelligence and large model proposed by SenseTime, and the autonomous driving that is heavy perception, large computing power, and does not rely on high-definition maps, it needs continuous technological innovation to promote.
This also means that intelligent driving in the next few years will also go more smoothly with the blessing of various technology companies and move towards a more mature industrial period.