Cars have taken different shapes in different eras. Nowadays, the car is no longer a simple means of transportation, and AI technology has become the "soul" of the car, making the car like a "Transformer", with the dual identity of a travel butler and a close friend. It can not only be like a smart and capable robot butler, autonomously complete all kinds of driving tasks, but also can incarnate the emotional assistant of the car owner, play games, watch dramas, chat freely, and bring the car owner a pleasant journey experience in different ways.
Based on Baidu's AI large model ecological empowerment, Jiyue Automobile Robot has ushered in a new evolution. On March 25, as the industry's first automotive AI technology conference, Jiyue held AI DAY 2024 in Beijing, and Xia Yiping, CEO of Jiyue, released the new version of OTA V1.4.0 software. Powered by world-class leading AI technology, Jiyue will set an industry benchmark in the era of AI travel.
V1.4.0 is also the 3rd consecutive launch of Jiyue 01 within 90 days and the 5th OTA after launch. This upgrade involves many fields such as intelligent driving, intelligent cockpit, intelligent interconnection, and three electrics, with a total of 200+ functions upgraded and 5300+ user feedback problems solved. For example, the visualization effect of OCC occupation network is upgraded, the two-way integration of SIMO and Wenxin Yiyan large model is integrated, and the human-vehicle-home ecology based on the large model is officially "on board". V1.4.0 will bring users a better, easier to use, and more fun smart travel experience, once again fulfilling the promise of allowing users to "drive new cars every month".
Baidu's AI large model ecology is deeply empowered, and the evolution of Jiyue automotive robots is accelerated
The automotive revolution has experienced the "fuel vehicle 1.0 era" and the "electric vehicle 2.0 era", and in the current "intelligent car 3.0 era", the underlying driving force comes from AI large model technology. The deep integration of AI models and the automobile manufacturing industry will lead the revolution of a new generation of automotive products.
The era of smart cars 3.0 is the era of automotive robots, and the starting point of the era change is the transfer of the "driving rights" of automobiles from humans to AI. Jiyue Automotive Robot clearly defines the product characteristics of the 3.0 era, which should meet the core capabilities of "three self".
(Jiyue CEO Xia Yiping)
Since the birth of Jiyue Automotive Robot, the core product concept of "natural communication, free movement, and self-growth" has been taken to enable smart cars to think and act like people, and to continue to evolve. Xia Yiping said: "In the era of smart car 3.0, 'robotization' will be the development trend, and the car with the three capabilities of natural communication, free movement and self-growth is the car robot, which is the real 'AI CAR'." Jiyue Automobile Robot allows everyone to see the true appearance of smart cars, and feel the technological innovation, efficiency improvement, and experience subversion brought by AI. ”
As a high-end intelligent car robot brand, Jiyue Automotive Robot has Geely SEA's vast architecture and Baidu's AI capability ecological empowerment, and has top-notch mechanical literacy and the industry's first-class intelligence level. Thanks to the ecological support of Baidu's AI large model, Jiyue 01 has deeply integrated technical capabilities such as Wenxin Yiyan, Apollo, Baidu Map, and Xiaodu. Jiyue 01 has created many industry leaders such as the world's first AI large model "onboarding", the only one in China/the only one in the world for pure vision high-end intelligent driving, the first OCC occupancy network upgrade, and the first full localization of AI voice algorithms.
In this AI DAY, Baidu's three latest AI native technologies in the fields of autonomous driving, LD maps, and human-computer interaction were unveiled, and will be applied to Jiyue automotive robots.
In terms of high-end intelligent driving, based on Baidu's Apollo pure vision high-end intelligent driving capabilities and safety system empowerment, Jiyue has completed the OCC occupation network upgrade and has formed a complete technical system of "B.O.T". At this AI DAY, Baidu Apollo autonomous driving vision model VTA (Vision Takes All) was officially released, which has greatly upgraded its capabilities including dynamic and static detection, timing tracking, real-time mapping, scene understanding, etc., and once again increased the upper limit of Jiyue's pure vision high-end intelligent driving ability.
(Wang Liang, Chief R&D Architect of Baidu Intelligent Driving Business Group and Chairman of IDG Technical Committee)
Wang Liang, Chief R&D Architect of Baidu's Intelligent Driving Business Group and Chairman of the IDG Technical Committee, said: "Based on the large model, Baidu has built the industry's first intelligent driving data production line and LLM-enabled autonomous driving data index, and at the same time, through generative AI technology, Baidu also has the ability to efficiently process long-tail data, which are important data engines to promote the development of end-to-end autonomous driving technology." ”
Regarding intelligent driving maps, "with maps" and "without maps" have become hot topics in current intelligent driving. Baidu Maps LD map (LD is lane-level navigation) is a native map of autonomous driving, which is generated by the large visual perception model of autonomous driving, completely reconstructs the traditional map production mode, realizes a new paradigm of end-to-end map generation, and breaks through the bottleneck of large-scale lane-level maps. LD is born for autonomous driving, which minimizes the requirements for the elements and accuracy of traditional high-precision maps, retains the most necessary elements and accuracy, and adds elements that traditional high-precision maps do not have, such as experience layers, safety layers, and real-time layers, so that autonomous driving can be driven more safely.
(Shang Guobin, Vice President of Baidu)
Jiyue 01 fully integrates the core capabilities of Baidu maps, as long as Baidu map navigation can reach the place, Jiyue 01's PPA intelligent driving can be used. Shang Guobin, vice president of Baidu, said: "The LD map has the three characteristics of 'all cities across the country', 'mobile phones/cabins can be used', and 'intelligent driving is safe with pictures', with full coverage of 360 cities across the country, and the only one in the world that applies the native map of autonomous driving to mobile phones and cars, so that intelligent driving can be driven like a person and safer than a person." ”
At the level of voice interaction, in order to make car robots smarter and more responsive in the process of interacting with users. Based on the underlying capabilities of Baidu's voice interaction, Jiyue 01 has built the voice assistant SIMO into a voice ceiling with the "global full-end side" of in-vehicle voice, with a recognition speed as fast as 500ms, and smooth interaction in weak/no network environments. In the future, the multi-modal interaction solution of video-voice fusion will be launched soon, comprehensively improving the interaction capability of multi-modal fusion and user privacy protection, and further improving the accuracy of speech recognition through the all-in-one device-side model.
（百度语音首席架构师贾磊）
Jia Lei, chief architect of Baidu Voice, said: "The more powerful technology must be the device-side large model model, so that the car machine can quickly respond to user interaction needs and answer questions about vehicle-related operations in any network environment, bringing users a perfect voice interaction experience on the device side." Video-voice fusion, all-in-one, and device-side large models are all industry-leading revolutionary technologies, all of which will be used to support the interactive upgrade of Jiyue automotive robots. ”
PPA added Guangzhou Kaecheng, OCC 3D map to visualize perception and recognition
The innovation of software will bring more surprises to smart cars. Automotive robots, which have been focusing on rapid upgrades and iterations, will bring an interesting "soul" to consumers.
As an important part of the Extreme Vision Intelligent Driving Solution, the OCC occupation network has been upgraded again, and the perception ability has been greatly improved. On the basis of the centimeter-level 3D model depiction comparable to LiDAR, the types of obstacle recognition have increased again, including not only individual static obstacles such as construction signs, fences, barricades, and anti-collision barrels, but also obstacles such as temporary road construction fences, faulty vehicles parked on the side, and temporarily stacked large garbage cans, so as to once again improve the adaptability of high-end intelligent driving in the process of driving.
It is worth mentioning that in order to allow users to better perceive the principle of OCC technology and use it with more peace of mind, V1.4.0 adds OCC 3D map easter eggs, which allow users to clearly see the world in the eyes of the car robot through the car screen. Blocks of different colors like building blocks identify obstacles, green belts, and other vehicles outside the car, and cartoon 3D characters identify passers-by, adding a lot of fun to the obscure AI technology. The function also supports turning on and off by voice, which is very convenient to control.
Thanks to the improvement of OCC perception capabilities, the point-to-point pilot assistance PPA has also been greatly upgraded, which can cope with more complex driving scenarios. For example, at an intersection with complex road conditions, where the car in front turns too slowly and the battery car is too close, PPA can actively detour to improve efficiency. In the event of sudden road construction fences, roadside faulty vehicles, temporary stacking of debris, etc., PPA can flexibly bypass or brake in time. In view of the problem that many users miss the exit of the high-speed ramp because of long solid lines or large vehicles blocking the road, PPA can also reasonably plan the route for them, find the right time to change lanes, so that every Jiyue user can avoid taking the wrong road.
In addition, the city PPA has opened a new city - Guangzhou, which is also the fifth city after Beijing, Shanghai, Hangzhou and Shenzhen. For the more special urban sections of Beijing, the PPA has also been further improved and the coverage has been expanded, which can fully cover the main roads in the inner urban area of the Fifth Ring Road.
At present, nearly 90% of users are using Jiyue PPA intelligent driving, and the cumulative mileage has exceeded 1.8 million kilometers, which is equivalent to 45 times around the earth. Next, the coverage mileage of PPA intelligent driving will exceed 500,000 kilometers. In 2024, based on the AI large model capabilities of Baidu Apollo and Baidu Maps, the goal of "being able to drive the whole country and having a good road experience" is just around the corner.
Jiyue 01 is the world's first smart car that supports voice control outside the car/parking, and the AVP and automatic parking assistance APA functions of valet parking are industry-leading, of which the APA penetration rate is 98%, which can be called the strongest parking on the surface. In the V1.4.0 version, the parking experience has been greatly upgraded, with the addition of the function of searching for space parking spaces in open roads, the addition of active avoidance of oncoming vehicles during parking, and automatic locking after parking, etc., with a level of care and safety awareness comparable to that of a five-star parking staff.
V1.4.0 also upgrades a series of driving experience optimizations, such as the addition of automatic turn signal function, the addition of PPA high-speed elevated lane change style settings, the addition of blind spot warning for automatic steering image assistance, and the optimization of ADAS functions such as Forward Collision Warning (FCW) and Automatic Emergency Braking (AEB). In the future, the choice of urban PPA driving style will also be updated.
The interaction is more natural and the entertainment is more sensitive, and the human-car ecology based on the large model is officially on the car
Nowadays, more and more people want the car to be like a "mobile living room", with a variety of functions to meet the needs of almost every family member.
Thanks to the underlying capabilities of Baidu's AI model, SIMO has achieved two-way integration with Baidu Wenxin Yiyan, bringing a more natural, smooth and logical human-computer interaction experience. In V1.4.0, SIMO is transformed into an all-round butler, adding active recommendation based on large models and free Q&A of car knowledge, supporting functions such as music recommendation, video information Q&A, scenic spot recommendation and query, and function answers. If you have any questions, ask SIMO, and the level of natural communication has reached a higher level, forming a significant generation gap with the stereotyped and mechanical voice assistants on the market.
For example, if you talk to a friend in the car about a TV episode but can't remember the details, you can quickly find the plot and play the episode by asking SIMO. Various questions about how to use PPA, such as smart driving, can also be easily answered through SIMO.
In addition, a new function has been added to share the content generated by large models. SIMO supports one-click sending of generated text, pictures and other content to the Jiyue APP, which is convenient for users to view and share. The SIMO experience outside the car has been further optimized, with the addition of continuous voice dialogue outside the car and improved recognition accuracy.
SIMO's capabilities have become the benchmark in the industry. At present, SIMO's average daily voice interaction penetration rate is 98%, and the average number of voice interactions per day is as high as 63 times, which is 10 times that of industry competitors. Among them, each vehicle invokes Wenxin Yiyan service 7.1 times per day, accounting for about 11% of the average number of voice interactions per day.
Rapid iteration and efficient closed-loop allow automotive robots to continue to evolve. Since the launch of Jiyue 01 for 5 months, 23,323 user feedback has been received through SIMO, and each problem will be connected to the R&D center with second-level efficiency, and users will be fed back within 24 hours, so as to achieve rapid upgrade and iteration.
In terms of people-car-home interconnection, which is of general concern to users, Jiyue 01 has been interconnected with the small ecological network, which means that the human-car-home ecology based on the large model has officially landed. Users can complete car condition inquiry through small speakers at home、Remote vehicle control and abnormal event reminder and other operations,Grasp the real-time status of the car at will。
A more user-friendly upgrade is that the V1.4.0 version also adds a new "All-Weather Extreme Mode", which has a wider range of use and more powerful functions than the general camping mode. Turning on the all-weather extreme mode can keep the vehicle powered on at all times, and functions such as air conditioning and entertainment services can be used normally, especially suitable for camping, sleeping in the car, temporarily leaving pets in the car and other scenarios, so as to bring users a more comfortable travel and life experience.
Multimedia entertainment ecosystem services have always been a high-frequency function for users. V1.4.0 adds a wireless screen projection app that supports screen projection to the in-vehicle device and an in-vehicle Wi-Fi hotspot for in-car devices to share traffic. At the same time, the games "All-Star Street Ball Party" and "Little Triangle Big Hero" were also launched.
In addition, it also includes the addition of a new app store, a new Baidu network disk, a new App remote car control and other functions, such as charging experience, steering wheel feel, sentry mode, audio experience, etc. have also been optimized and upgraded. Whether it's coming home from work, picking up people or going out camping, Jiyue car owners can spend a fulfilling, fun and comfortable time in the car.
Xia Yiping announced that the Jiyue V2.0 version will be released soon, and it will be upgraded with a new voice model, PPA intelligent driving and other blockbuster functions that can be opened nationwide.
Looking to the future, AI technology will surely create a smart world that subverts imagination. Based on the ecological capabilities of Baidu's AI large model, Jiyue will bring users a more intelligent and comfortable life of automotive robots, and will also help accelerate the evolution of the industry and lead the era of automotive robots.