Zebra Zhixing, a joint venture between Alibaba and SAIC, has been diligently working on the advancement of smart cabin technologies. The latest innovation involves a dual-model architecture labeled “large model + small model,” which effectively addresses the technological challenges of achieving both deep intent understanding and real-time response within the smart cabin environment. Traditionally, smart cabin systems were limited by the performance of single models, which could either fail to fully understand the complex demands of users or struggle to deliver responses quickly enough. What Zebra Zhixing has achieved with this innovation is a compelling resolution to this longstanding dilemma.

The newly enhanced Yuanshen AI system has revealed three critical advancements. Firstly, it offers open-ended conversational interactions based on deep logical reasoning. Previous voice interaction systems within smart cabins were largely limited to simple command-response scenarios, where elaborate user inquiries often went unanswered or misunderstood. In contrast, the integrated Yuanshen AI system, buoyed by the robust logical reasoning capabilities of the DeepSeek large model, allows for more natural and profound dialogue with users, interpreting the deeper meanings behind their words and providing targeted responses. For instance, if a user asks, “What should I prepare for my beach outing tomorrow?” the system not only grasps the user's travel intent but can also suggest items such as sunscreen, sunglasses, and swimwear while even recommending nearby beach attractions and travel tips.

Secondly, the AI has been equipped with powerful multimodal understanding capabilities to accurately identify vague user demands. In the context of smart cabins, user requests may manifest in diverse forms such as voice, gesture, or even eye movements, which can sometimes be ambiguous. The Yuanshen AI System excels in aggregating and analyzing various inputs—be it vocal commands, hand movements, or facial expressions—to discern user intentions. For example, if a user gestures toward the car window while saying, “It’s too sunny,” the system can quickly comprehend the user's desire for shading and automatically adjust the sunshade accordingly.

Thirdly, this innovative setup introduces a proactive service model that enables connectivity between the vehicle control systems and ecological services, thereby facilitating intelligent combinations of service recommendations—be it for dining options or travel itineraries. This marks a significant shift from reactive to proactive interactions, wherein the smart cabin can sense the user’s context and requirements, preemptively delivering services. For instance, should the system recognize that it is lunchtime and the user is in an unfamiliar area, it would autonomously recommend suitable restaurants based on the user's taste preferences while also mapping out the best route to get there. In planning a long-distance drive, the system could seamlessly compile information about charging stations, rest areas, and weather conditions, providing comprehensive travel advice. This platform has now become the industry’s first AI solution in smart cabins deeply integrated with DeepSeek capabilities, establishing a new technological benchmark.

According to CTO Si Luo, the architecture behind the Yuanshen AI embodies a layered processing mechanism that successfully combines the complex computations of the DeepSeek large model with the rapid responses of lightweight small models. Within this architecture, the DeepSeek large model handles intricate tasks pertaining to semantic comprehension, logical reasoning, and knowledge retrieval to furnish the system with robust intelligent support. Meanwhile, the lightweight small model addresses the swift execution of basic commands such as simple vehicle controls and information inquiries. This innovative structure not only meets the deep understanding requirements of cabin interactions but also reduces service response times by 60% through local deployment, effectively tackling the common lag seen in large model applications within vehicular environments. The local deployment facilitates the rapid processing of vast data sets directly in the vehicle, minimizing delays associated with data transmission and cloud computing, thereby significantly enhancing the user experience.

Leveraging foundational technology from AliOS, Zebra Zhixing has established a vast cooperative ecosystem encompassing over 30 leading automotive brands worldwide. Recent statistics reveal that as of November 2023, their smart automotive solutions have surpassed an installation milestone of three million units, collaborating with renowned car manufacturers including SAIC, FAW, BMW, Volkswagen, and Changan. With its cutting-edge technology and premium services, Zebra Zhixing has gained the trust and partnerships of numerous automakers. This technological upgrade is set to further solidify Zebra Zhixing's position at the forefront of smart automotive solutions, propelling the transition of smart cabins from mere “functional integration” to comprehensive “scenario-based services.” As smart cabin technology continues to evolve, it is anticipated that Zebra Zhixing will create even more groundbreaking innovations in the automotive space, providing users with a smarter, more convenient, and comfortable travel experience.