The 'China Intelligent Connected Vehicle Assessment Protocol (C-ICAP) 2026 Edition' has been officially implemented on July 1, 2026. Unlike C-NCAP, which evaluates 'safety,' C-ICAP focuses on 'intelligence'—independently rating four units: driving assistance, parking assistance, smart cockpit, and privacy protection. This article provides a detailed breakdown based on the original management rules and Appendices A-D.
01 Overall Framework: Four Major Units and Three-Tier Evaluation
C-ICAP assigns star ratings to each of the four units individually, with star rating thresholds consistent with C-NCAP: a comprehensive score rate of ≥92% for 5+, ≥83% for five stars, and ≥74% for four stars.
- Driving Assistance (Appendix A, 119 pages): Evaluates the actual performance of L2/L2+ combined driving assistance on highways and in urban areas.
- Parking Assistance (Appendix B, 71 pages): Evaluates the full capability from APA (Automated Parking Assist) to AVP (Automated Valet Parking).
- Smart Cockpit (Appendix C, 154 pages): Evaluates cockpit interactions such as voice, touch, surround view, HUD (Head-Up Display), and sentry mode.
- Privacy Protection (Appendix D, 52 pages): Evaluates IoV (Internet of Vehicles) security and personal information protection.
Each unit is internally divided into three tiers: on-site test items (real-vehicle testing by China Automotive Test Center), report review items (third-party test reports submitted by enterprises), and bonus/deduction items.
02 Driving Assistance: 93% Weight on Highway and Expressway Scenarios
This is the most noteworthy weight distribution in C-ICAP: among the primary indicators for driving assistance, highway and expressway scenarios account for 93%, urban scenarios for only 5%, and human-machine interaction for 2%. This means that the 'driving assistance' evaluated in the 2026 edition of C-ICAP is essentially highway NOA (Navigate on Autopilot) capability, with urban NOA merely being the icing on the cake.
2.1 Highway and Expressway Scenarios (93% Weight)
The highway and expressway scenarios include four secondary indicators:
- Main road passing capability (40%): Encountering stationary targets ahead (motorcycles + passenger cars, fallen cones, fallen tires), slow-moving targets, decelerating targets, target cut-in and cut-out, and false activations (must not falsely brake when overtaking a sprinkler truck).
- Ramp passing capability (10%): Success rate of merging into the ramp, success rate of merging into the main road, divided into daytime/nighttime.
- Complex environment passing capability (25%): Accident areas (rollover vehicles + triangular warning signs), road construction, stationary motorcycles/passenger cars in tunnels, rollover vehicles + pedestrians with umbrellas in rain, and construction areas in fog.
- Comprehensive continuous operation capability (25%): Frequency of system intervention, frequency of system exit, lane change success rate, activated mileage proportion, passing efficiency, lateral/longitudinal control comfort, service area passing, and toll station passing (bonus item).
2.2 Newly Added Highway and Expressway Scenarios in the 2026 Edition
The preface of the management rules clearly states that the 2026 edition adds several scenarios with extremely high requirements for perception systems:
- Irregular targets: Fallen tires, fallen cones, rollover vehicles—these are not standard vehicle targets and are difficult to identify relying solely on millimeter-wave radar, posing a direct test for visual perception.
- Multiple targets: Multiple targets appear ahead simultaneously, requiring the system to determine which one is the real risk.
- Tunnel scenarios: Sudden light changes and GPS signal loss inside tunnels test the localization capability of vision + map fusion. Rain and fog weather: Simulates rainfall and dense fog environments where camera performance degrades.
- Toll station/service area passing: Automatically entering and exiting toll stations and service areas via NOA, which is the 'last mile' of highway NOA.
2.3 Urban Scenarios (5% Weight, Review Items)
Although urban scenarios account for only 5% of the weight, the scenario design is highly detailed: encountering stationary targets ahead (sprinkler trucks on curves, shared bicycles on curves, squatting children, dogs), slow-moving e-bikes, target crossing (occluded target groups, diagonal crossing of shared bicycles, dog crossing, reversing passenger cars with occlusion), traffic light recognition and response, intersection conflicts (left turn encountering target groups in the same direction, going straight encountering occluded distant motorcycles running red lights, right turn encountering illegally parked vehicles), main and auxiliary road intersection conflicts, and roundabout passing.
Newly added scenarios include passing through variable lanes and tidal lanes (bonus items). These are all complex working conditions unique to Chinese urban roads.
2.4 Human-Machine Interaction (2% Weight)
The 2026 edition has made major adjustments to human-machine interaction: Driver Monitoring System (DMS) has been removed, and three new categories of items have been added:
- Physical buttons: Whether there are physical buttons for activation/deactivation/speed adjustment methods; it cannot rely entirely on the touch screen.
- System deception: Hand-off detection system deception—using a fake hand or heavy object on the steering wheel to simulate 'hands on the steering wheel'; the system must be able to identify this deceptive behavior and exit.
- Occupant posture monitoring (bonus item): The system must be able to identify and alert if the occupant's head is too close to the airbag or if the occupant's feet are placed on the dashboard.
Traffic violations are deduction items: speeding, occupying the emergency lane, driving in the wrong direction, and crossing solid lines; points are deducted for each occurrence.
03 Parking Assistance: 65% Focus on Parking In, AVP Accounts for Only 6%
Primary indicator weights for parking assistance: parking-in assistance 65%, parking-out assistance 20%, human-machine interaction 9%, indoor cruising 3%, and outdoor cruising 3%. Parking-in/parking-out/human-machine interaction are mandatory test items, indoor and outdoor cruising (i.e., AVP) are review items, and remote control parking is a bonus item.
3.1 Parking-In Assistance (65%)
- Parallel parking-in (45%): Narrow spaces, vehicle parked at a 6° angle, controllable area 3.6m wide, parked vehicles ahead, dynamic interference during parking-in, occupied parking spaces, and nighttime parking-in.
- Perpendicular parking-in (40%): Narrow spaces, single-side pillar obstacles, dynamic interference, occupied parking spaces, parking-in after other vehicles leave, nighttime parking-in, encountering adjacent car doors opening during parking-in, and parking-in conflicts.
- Angled parking-in (15%): Narrow spaces and standard parking spaces.
New key additions in the 2026 edition: nighttime parking-in (20% weight), dynamic interference during parking-in (21-23% weight), and adjacent cars suddenly opening doors—these are all the most common failure scenarios in real parking lots.
3.2 Parking-Out Assistance (20%)
Parallel parking-out and perpendicular parking-out each account for 50%. New additions in the 2026 edition: parked e-bikes in front of the parking space (e-bikes are narrower than cars and are easily ignored by the system), and rear-end parking-out methods.
3.3 Indoor/Outdoor Cruising Capability (3% Each, Review Items)
This is the assessment for AVP. Indoor cruising evaluates mapping efficiency, cruising efficiency, encountering fallen construction tools, encountering baby strollers, encountering closed construction areas, encountering children crossing diagonally, encountering continuous pedestrians, meeting vehicles at intersections, and starting/stopping on slopes.
New additions for outdoor cruising: cardboard boxes, stone piers, fallen bicycles, children crossing with occlusion at night, slow-moving oncoming pedestrians at night, front vehicle parking-out, and narrow passages at night.
The squatting child target is a new addition in the 2026 edition—children squatting and playing in parking lots have a very low profile, making them difficult for sensors in ordinary parking systems to detect.
04 Smart Cockpit: Voice Interaction Accounts for 60% of Intelligent Interaction
The smart cockpit is divided into test items (85%) and review items (15%). Primary indicators for test items: intelligent interaction 55%, intelligent escort 30%, and smart services 15%.
4.1 Intelligent Interaction (55%)
- Voice interaction (60%): Wake-up success rate (no noise 30% + noise 50% + false wake-up 10% + outside-car wake-up 10%), wake-up time, interaction success rate (no noise 30% + noise 50% + offline interaction), interaction response time, and voice vehicle control safety (deduction item—deducting points for using voice to control safety-related functions such as car lights/gears/wipers while driving).
- Touch interaction (40%): Display quality, response time (answering calls/turning on AC/navigation/music/video each 20%), sliding fluency (contacts/music list/navigation overview), operation convenience (AC/radio/audio source switching), touch operation safety (deduction item—touching car lights/gears/emergency calls/wipers/honking while driving), and key symbol display (deduction item—whether high beam/fog light/low beam/ADAS icons are clear).
Review items include: AI voice interaction (travel assistant/car butler/entertainment recommendation/encyclopedia/fun chat/image and text generation), multi-screen linkage (video/navigation/music linkage), and multimodal interaction.
4.2 Intelligent Escort (30%)
- Surround view (60%): Activation methods (low-speed steering/stalk button voice/obstacle too close), activation time, image display delay (rear camera delay weight 30%), and display quality (daytime/nighttime).
- Anti-glare (40%): Glare degree of the central control screen on the main interface/navigation/music interface; ghost images are a deduction item.
Review items include: Head-Up Display (HUD) and sentry mode.
4.3 Smart Services (15%)
- Smart unlocking (45%): Unlock/lock success rate (short-range 50% + remote 40% + physical key 10%), seamless unlocking stability (unlock distance stability 70% + abnormal unlock/lock 30%), and remote vehicle control success rate (windows 20% + AC 30% + smart car finding 30% + vehicle status 20%).
- In-vehicle navigation (30%): Lane prompt accuracy (double fork 30% + multiple forks 30% + tunnel 40%), lane localization response time, route guidance accuracy, route guidance response time, and reminder functions (traffic light countdown/start reminder, etc.).
- Phone-vehicle interconnection (15%): Bluetooth connection success rate (regular calls 40% + WeChat calls), WiFi interconnection, phone ecosystem flow (navigation/video/music), and phone wireless charging.
- Smart accessories (10%): In-vehicle software system (account login/software ecosystem/external expansion) and smart seats (ventilation heating/massage/position memory).
Review items also include smart audio effects (sound quality, adaptive sweet spot).
05 Privacy Protection: From Bluetooth Attacks to Adversarial Machine Learning
Privacy protection is divided into test items (85%) and review items (15%), with two primary indicators: connected communication protection and personal information protection.
5.1 Connected Communication Protection
- Basic security protection: In-vehicle Bluetooth (fuzz testing/secure pairing/access control/Bluetooth key spoofing attacks), in-vehicle WiFi (Sybil attacks/AP-side security authentication/password rules), NFC keys (fuzz testing/random number detection), RF remote control keys (fuzz testing), and vehicle control APP remote unlocking (client program security/sensitive information security).
- Navigation positioning spoofing (new in 2026): BeiDou BDS positioning spoofing, GPS positioning spoofing—attackers send fake satellite signals to make vehicle navigation deviate from the route.
- Software upgrades (review item): OTA upgrade security, offline upgrade security. Whole-vehicle known vulnerability detection (review item): Bluetooth/WiFi/vehicle control APP known vulnerability database detection. Whole-vehicle in-depth security testing: freely combined attack use cases.
- Adversarial attacks on combined driving assistance systems (new in 2026): Adversarial example attacks on perception systems (cameras/radar)—such as pasting stickers on traffic signs to make the system recognize a 60 km/h speed limit instead of 80 km/h.
- Adversarial attacks on intelligent voice models (new in 2026): Generative AI security—using adversarial voice commands to control the in-vehicle system without the user hearing it.
5.2 Personal Information Protection
- Sensitive personal information on the vehicle side: In-vehicle video and images (user notification/collection authorization/collection prompt/access authorization/data deletion), in-vehicle audio (facial recognition/user notification/collection authorization), vehicle location information, web browsing records, financial payment records (new in 2026—local desensitization/access authorization), outside-vehicle video and image information (new in 2026—anonymization detection effect of sentry mode), and privacy security mechanisms (new in 2026).
- General personal information on the vehicle side: Personal mobile phone number (transmission encryption/local desensitization), user personal account, license plate number, and audio/video browsing records.
- Personal information protection on the mobile phone side: Vehicle control mobile APP—excessive information collection, unauthorized third-party sharing, unauthorized plugin embedding, unauthorized personal information collection, and APP secondary verification.
06 Five Judgments for R&D Teams
First, highway NOA is the main battlefield of C-ICAP. With 93% of the weight placed on highway and expressway scenarios, it means that no matter how well urban NOA is performed, a poor highway NOA will fail to achieve five stars. R&D resource allocation should prioritize ensuring the robustness of highway scenarios—irregular target recognition, tunnel localization, rain and fog weather perception, and ramp merging success rate.
Second, parking tests the 'success rate under extreme working conditions.' 65% of the weight is on parking-in, with nighttime parking-in and dynamic interference each accounting for over 20%. This means that successfully parking in a standard space during the day is not impressive; the real skill lies in being able to park in at night, when adjacent cars open their doors, or when people walk by.
Third, voice interaction is the core battlefield of the cockpit. Intelligent interaction accounts for 55% of the cockpit's weight, of which voice interaction accounts for 60%. The weight of interaction success rate in noisy environments is 50%—whether voice can still be accurately recognized in real-world scenarios such as driving at high speeds with windows open, maximum AC airflow, and passengers chatting directly determines the cockpit's star rating.
Fourth, privacy protection has shifted from 'compliance' to 'offensive and defensive confrontation.' The 2026 edition adds positioning spoofing, perception adversarial attacks, and voice model adversarial attacks—this is no longer a compliance question of 'whether user notification has been done,' but a technical question of 'whether real hacker attacks can be blocked.' Security teams need to shift from passive compliance to proactive offense and defense.
Fifth, physical buttons and system deception are new deduction items. Although human-machine interaction accounts for only 2% of the weight, the two items of 'hand-off detection deception' and 'touch safety while driving' directly correspond to recent industry hot topics—deceiving DMS with fake hands and touch screen operation safety while driving. By including these in the assessment, C-ICAP is guiding the industry to return to the product logic of 'safety first.'