Automotive AIOS Research: Mass Production Solutions Are Implemented
Mass Production Solutions Are Implemented on A Small Scale.
In 2026, AIOS starts small-scale implementation, helping to improve various cockpit AI functions and enable more comprehensive application scenarios. In addition, the AIOS of some mainstream flagship vehicle models realizes cross-domain orchestration capabilities through atomic services, expanding execution capabilities to body, chassis, intelligent driving and other domains.
As of June 2026, OEMs have still adopted two AIOS R&D models: self-development and semi-outsourcing:
Full-stack self-development: Emerging automakers led by NIO and Li Auto deeply integrate AI capabilities into the middleware layer (even kernel layer), forming a full-stack closed loop from chip to application.
Semi-outsourcing: Traditional OEMs independently build the vertical large model + AIOS framework, reusing basic software from suppliers at the bottom layer.
The in-vehicle deployment modes of mass-produced AIOS solutions of suppliers include the following:
Extending from cockpit AI applications to the bottom layer of OS: the mainstream approach. For example, Huawei provides HarmonyOS-related services via HarmonySpace.
Binding with chip/hardware manufacturers: Adapt to chip solutions of multiple chip manufacturers for software-hardware collaboration. Typical examples include Sage Box for SenseTime SageOS and AI Box-N1 for ThunderSoft AquaDrive OS, which build comprehensive on-device AI solutions coordinated with AI Box.
Binding with cloud providers: Represented by Extour Technology, bind with Volcano Engine’s cloud base and invokes Doubao Large Model to provide AI services.
Compared with 2025, cross-domain invocation services of OS became more mature in 2026.
In the case of Dongfeng Tianyuan OS, the entire architecture realizes integration across five domains: body, powertrain, chassis, thermal management and gateway. Based on the Taichi Large Model base, it invokes more than 2,000 atomic services and supports rapid combination and flexible invocation of functions via service-oriented architecture.
In the process of AIOS deployment, the software foundation for cross-domain invocation of cockpit software system is still the vehicle OS. Based on the vehicle OS, non-safety cockpit functions can be disassembled atomically. Then, automotive intelligent scheduling algorithms carried by AI middleware realize dynamic allocation and intelligent scheduling of computing power, applications and peripheral resources in the cockpit. When the user issues an instruction, the voice assistant disassembles intentions, coordinates multiple AI frameworks to work, and finally invokes atomic services. This entire process is the basic workflow of the AIOS.
In 2026, the number of interfaces for atomic capabilities surged (mainstream flagship vehicle models generally have more than 500 atomic capabilities). Against the backdrop of increasingly popular customized cockpit scenarios, the competitive edges of AIOS have gradually shifted from "more atomic capabilities" to "easier combination of atomic capabilities". Protocols for customized and standardized interfaces are critical on the issue of “whether to combine more easily”.
Some of the engineering challenges involved in highlighting the effects of AIOS are as follows:
The lack of standardized unified protocols makes it very easy to hinder the effectiveness of atomic capabilities under SOA. At present, Function Call is the mainstream protocol adopted, while MCP is still in the trial stage. The reason is that Function Call can meet core requirements and is easy to maintenance under small-scale mass production conditions. However, when large-scale migration of solutions is required, wrapping MCP Server outside Function Call demonstrates advantages of "cross-model portability" and "dynamic tool discovery".
The significance of the MCP protocol lies in standardization, compressing the development cycle of cross-scenario functions from months to weeks. Automakers can quickly combine personalized cockpit services like building blocks. Typical cases include Extour Technology's automotive MCP-Agent framework and SenseAuto's edge native agent framework supporting MCP/A2A protocols.
For example, SenseAuto launched an edge native agent framework supporting MCP/A2A protocols. It builds a standardized "Agent-Tool" integration framework, allowing multiple agents to efficiently integrate various vehicle tools such as players, air conditioners and knowledge bases through a unified MCP protocol layer, solving difficulties in tool invocation, data acquisition and multi-source information integration during agent development.
Its advantages include:
Cost reduction and efficiency improvement: The unified protocol eliminates fragmentation barriers for tool docking, greatly cutting development and collaboration costs and enabling all types of tools to be "plug-and-play".
Open ecosystem: Supports a standardized ecosystem access mechanism, facilitating rapid integration of third-party services and hardware into intelligent vehicle systems, and promoting diversified ecosystems.
Controllable security: Unified security authentication policies and centralized management simplify processes while strengthening system security.
Next Stage: Shift from AI-Driven to AI-Native
Suppliers including Huawei and Neusoft divide the integration of AI and OS into three stages:
In 2025, most OEMs and suppliers built AI operating systems by deploying the AI framework at the middleware layer. Examples include XPeng’s deployment of cross-domain unified protocol middleware + on-device large model + atomic services, Great Wall Motor’s deployment of multi-model base and Agent management/operation framework at the middleware layer, and Neusoft Reach’s deployment of NeuSAR AI Framework on NeuSAR OS for rapid introduction of AI applications into vehicles.
In 2026, leading emerging OEMs and suppliers start deploying "AI-enhanced kernels" to build kernel-layer native AIOS. For instance, NIO leverages AI to improve the OS kernel’s ability to dynamically schedule resources according to scenarios; Huawei HarmonyOS kernel natively supports multi-modal understanding and personalized data understanding.
In addition, with the deployment of agent technology and upgrading of high-compute chips, the AIOS architecture has also undergone changes from the application layer to the bottom layer.
For example, NIO’s new SkyOS deeply integrates AI capabilities into the bottom layer of the operating system, replacing the traditional architectural paradigm. It realizes efficient end-cloud integrated collaboration, intelligent scheduling of heterogeneous computing power, and multi-agent collaboration, while improving system response speed, stability, data throughput and battery life. Its innovation lies in enhancing the kernel’s ability to dynamically schedule resources according to scenarios, including CPU (process and thread priority), memory management (allocation and recycling) and device sharing. It can boost system stability and response speed in high-load scenarios.
In the case of Huawei Qiankun OS, it contains a security isolation engine, AI-native kernel, UnifiedBus, acceleration engine and Bisheng Compiler, providing deterministic low-latency rapid response for upper-layer ADS algorithms. Its kernel is based on HarmonyOS kernel but deeply tailored and reconstructed for automotive scenarios to become an AI-native kernel, and can realize seamless flow of "vehicle-road-cloud-mobile phone".
Automotive AIOS Research Report, 2026
Automotive AIOS Research: Mass Production Solutions Are Implemented
Mass Production Solutions Are Implemented on A Small Scale.
In 2026, AIOS starts small-scale implementation, helping to improve v...
Automotive Telematics Service Provider (TSP) Research Report, 2026
TSP Research: Leading providers collectively turn to AI agents to provide all-scenario active services
Telematics Service Providers (TSPs) are the core hub of the telematics industry chain, connectin...
Automotive Smart Interior Research Report, 2026
Smart Interior Research: As Technologies like Interactive Starlight Headliner, Hidden Display and Surface Projection Are Launched, Automotive Interiors Become Ever More Intelligent
The Automotive Sma...
Research Report on AI Applications in Cockpits, 2026
AI Application in Cockpits: AI Services Become More Comprehensive, Convenient, and Refined.
In the first half of 2026, cockpit AI functions underwent initial upgrades across multiple dimensions, inc...
Software-Defined Vehicles in 2026: OEM Software Development and Supply Chain Deployment Strategy Research Report
Research on OEMs’ Software Strategies: R&D Focus, Development Strategies and Supplier Building Models of 30 OEMs
In this paper, we adopt a research framework covering 13 subsystems and 48 sub-di...
Passenger Car Chassis Domain Control and Chassis Cross-Domain Integration Research Report, 2026
Chassis Control Research: Mass Production of Full Chassis-by-Wire Solutions Starts
1. A Cluster of Full Chassis-by-Wire Solutions Make Their Debut, and EMB Enters Mass Production and Adoption for the...
Central Domain Control (Powertrain, Chassis, Body) and Motion Controller Research Report, 2026
Central Domain Control and Motion Control Research: XYZ Coordinated Control and Full X-by-Wire Actuation System
With the gradual penetration of L3+ autonomous driving, the chassis control system is ...
48V Low-voltage Power Distribution Network (PDN) Architecture and Supply Chain Panorama Research Report, 2026
Research on 48V Low-Voltage Power Distribution Network （PDN): An Active 48V Supply Chain, with Priority Deployment in High-Power Scenarios Such as Steer-by-Wire Chassis
The automotive 48V low-voltage...
AI-Defined Vehicle (AIDV) OEMs' Deployment Strategies Research Report, 2026
AIDV Research: Deployment Strategies of 22 OEMs
The AI-Defined Vehicle (AIDV) OEMs' Deployment Strategies Research Report, 2026, released by ResearchInChina, analyzes the AI deployment strategies of ...
OEMs’ Passenger Car Model Planning Research Report, 2026
Vehicle Model Planning Research: Chinese OEMs Launch Sub-Brands Intensively, While Multinational OEMs Apply the Brakes to Electrification Strategies
ResearchInChina released the OEMs’ Passenger Car M...
Autonomous Driving Simulation and World Model Research Report, 2026
Autonomous driving simulation research: "Simulation test + world model"-driven test system has become R&D infrastructure.
The "Autonomous Driving Simulation and World Model Research Report, 2026"...
Cockpit-Driving Integration Central Domain Controller SoC and AI Supercomputing Architecture Research Report, 2026
Cockpit-Driving integration and AI supercomputing research: The One Chip solution is rapidly installed in vehicles, and AI supercomputing architectures are moving towards full-domain integration.
AI ...
Intelligent Driving End-to-End Large Model Research Report, 2026
Research on Intelligent Driving Large Models: A Critical Period for Technological Competition and Paradigm Integration
As autonomous driving technology rapidly iterates from L2 to L3?L4, intelligent...
Automotive Digital Key Industry Trend Report, 2026
Digital Key Research: Automotive BLE, UWB and SLE Hardware Layout
The Automotive Digital Key Industry Trend Report, 2026, released by ResearchInChina, analyzes and predicts the digital key market, co...
Monthly Report on Automotive New Technology (May 2026)
UHD gaze technology, full-color LiDAR, UWB, etc. promote the upgrade of intelligent driving perception capabilities
This report is published once a month and is available for annual subscription.The...
In-Cabin Monitoring Systems (DMS, OMS, etc.) Research Report, 2026
In-Cabin Monitoring System Research: DMS to Become Mandatory in 2027, Expected to be Installed in Over 14 Million Vehicles
ResearchInChina released the In-Cabin Monitoring Systems (DMS, OMS, etc.) Re...
Automotive Service-Oriented Architecture (SOA) and Cross-Domain Middleware Industry Report, 2026
Research on automotive SOA and cross-domain middleware: The era of AI atomic services and AI cross-domain fusion agents is coming.
Automotive SOA evolves towards AI + full SOA servitization Driv...
Automotive Display, Center Console and Cluster Industry Report, 2026
Automotive Display Research: Multi-Screen Application Slows Down, While OLED and MiniLED Are Introduced in Vehicles Quickly
In 2026, automotive displays will no longer excessively pursue the number a...