At the EAC2026 conference on May 29, 2026, Saad Nawaz, a technical expert from AUDI AG, delivered a keynote speech titled "Evolution of Automotive mmWave Radar Architectures," which is a rare and high-quality report.
Combining the upgrade path of autonomous driving, Saad Nawaz analyzed the evolving requirements and development trajectory of in-vehicle radar products. Key topics included: common challenging scenarios for mmWave radar, countermeasures for these scenarios, architectural upgrade paths for forward-looking and surround-view radars, as well as hot topics such as the value of satellite radar, AI (Artificial Intelligence) algorithm applications, and associated challenges.
This article extracts and analyzes the core content. The download link for the original report can be found in the pinned comments, or you can contact the author for access (contact information is at the end of the article or via direct message).
01.Application Scenario Analysis
1. Evolution of Requirements
As the functions of ADAS (Advanced Driver Assistance Systems) and autonomous driving systems continue to expand, and radar processing capabilities and computing power steadily advance, the role undertaken by radar is also constantly evolving.
In L2 and below driving assistance functions (such as ACC (Adaptive Cruise Control) and AEB (Automatic Emergency Braking)), the primary role of mmWave radar is forward detection and longitudinal assistance. For some L2++ functions (such as Urban NOA), it is necessary to handle intersection traffic scenarios, identify VRUs (Vulnerable Road Users), and achieve more complex interactive functions.
L3 and L4 autonomous driving functions impose higher requirements on mmWave radar, covering multiple aspects including detection range, FOV (Field of View), elevation detection capability, angular resolution, and complex environment understanding. mmWave radar must be capable of handling congested/high-speed driving, environment modeling, and cut-in behavior recognition. It needs to perform full-environment perception and all-weather detection, and support sensor redundancy design to meet functional safety requirements.
2. Common Challenging Scenarios
In advanced autonomous driving, there are 8 common challenging application scenarios for mmWave radar.
1) High-speed approaching motorcycles from the rear
Motorcycles and other small targets have a small RCS (Radar Cross Section), high relative speed, and fast trajectory changes, making them difficult to detect and posing a significant challenge for stable target tracking.
2) Stationary trucks under bridges
Road infrastructure such as bridges generates strong reflections, which can easily trigger multipath effects and generate false targets, leading to missed detections of stationary trucks under bridges.
3) Protruding cargo from leading vehicles
The target outline exceeds the vehicle body, resulting in weak radar reflection signals and blurred features, which makes it prone to missed detections.
4) Vulnerable road users next to stationary vehicles
VRUs are adjacent to strongly reflecting objects, resulting in weak reflection signals that can be easily overwhelmed by the strong reflections, thus highly susceptible to missed detections.
5) Small obstacles within the drivable area
Small obstacles have a low RCS, coupled with clutter interference easily generated by surrounding objects, leading to missed detections.
6) Occluded pedestrians cutting into the ego vehicle's lane ("ghost probing")
Due to occlusion and multipath interference, the detection timing for suddenly appearing pedestrians is often too late.
7) Pedestrians immediately adjacent to the ego vehicle
The close proximity of pedestrians to the vehicle causes mutual reflection of target echoes, resulting in excessive interference information and leading to missed detections.
8) Road edge/curb recognition
Road edges and curbs have low heights, making them difficult to detect. The complex environment leads to dense clutter, which easily causes unstable detection.
02.From Application Scenarios to Radar Performance Requirements
The core radar performance requirements for the above 8 application scenarios are as follows:
| Scenario No. | Specific Scenario | Challenge | Core Radar Performance |
| 1 | High-speed approaching motorcycles from the rear | Long-distance small RCS, high relative speed | Velocity resolution |
| 2 | Stationary trucks under bridges | Strong static reflection, prone to false/missed detections | Elevation angular resolution |
| 3 | Protruding cargo from leading vehicles | Target exceeds outline range, weak radar signal | Dynamic range / Detection sensitivity |
| 4 | Vulnerable road users next to stationary vehicles | Adjacent to strong reflecting objects, weak reflection signal | Azimuth angular resolution, Dynamic range / Detection sensitivity |
| 5 | Small obstacles within the drivable area | Low RCS, strong reflectors in the vicinity | Elevation angular resolution |
| 6 | Occluded pedestrians cutting into the ego lane | Short reaction time, strong reflectors in the vicinity | System latency |
| 7 | Pedestrians immediately adjacent to the ego vehicle | Ultra-close-range detection, severe signal interference | Near-field performance |
| 8 | Road edge/curb recognition | Low target elevation angle, strong environmental clutter | Elevation angular resolution |
In summary, the main aspects include detection sensitivity, angular resolution, dynamic range, near-field performance, and system latency.
1. Detection Sensitivity
The detection sensitivity design of traditional radars centers on the center of the FOV, with peak performance entirely concentrated in this area. This is because its original design primarily served longitudinal control functions in highway scenarios, resulting in insufficient attention to targets at the edge of the FOV with partial occlusion. This directly leads to generally low detection reliability of radar in the edge areas of the FOV.
As autonomous driving scenarios extend to complex urban roads, it is required not only that the radar achieves uniform and consistent detection performance across the full FOV and can stably identify small RCS targets at the edge of the FOV, but also that it maintains stable operation in dense urban traffic environments. Meanwhile, it must adapt to complex terrains such as slopes and urban sections with significant elevation differences, and reliably identify weakly reflecting targets at long distances and large elevation angles.
To meet the above requirements, radar design needs systematic optimization from multiple technical dimensions: increasing the antenna aperture, optimizing the antenna illumination method, adopting advanced beamforming technology across the full-angle area, and reducing sidelobe interference. Meanwhile, targeted gain compensation is performed according to different detection angles to achieve consistent dynamic perception capabilities at the center and edges of the FOV.
2. Azimuth Angular Resolution and Dynamic Range
High azimuth angular resolution and large dynamic range are the foundation for reliable perception in dense traffic conditions.
The azimuth resolution and dynamic range design of traditional radars generally only meet the basic target separation requirements in low-density highway environments. Their azimuth angular resolution is about 7°~10°, and the dynamic range is about 35~40dB. The performance metrics are mainly set around simple high-speed driving scenarios and cannot cope with the refined perception requirements in complex traffic conditions.
As autonomous driving technology evolves towards complex urban scenarios, the radar not only needs to accurately distinguish between different types of targets at close range, such as vehicles, motorcycles, and VRUs, and clearly parse dense urban traffic conditions, but also reliably identify weak targets next to strong reflectors (such as pedestrians next to trucks), while possessing excellent anti-interference capabilities. In terms of specific metrics, the azimuth angular resolution is required to be less than 0.5° at a distance of 100 meters, and the dynamic range must be no less than 70dB.
To match the above performance requirements, the radar needs comprehensive upgrades from the hardware architecture to the algorithm level: ensuring high angular resolution by increasing the antenna aperture, significantly boosting the number of virtual channels by adopting advanced beamforming and MIMO (Multiple Input Multiple Output) technology, optimizing the dynamic range signal processing chain, and leveraging AI and machine learning algorithms to accurately extract weak target signals from complex background clutter.
3. Elevation Angular Resolution
Elevation angular resolution enables true 3D spatial perception in complex environments.
The elevation perception capability of traditional radars has fundamental limitations, with limited vertical discrimination ability, low elevation angle estimation accuracy, and very weak height classification capability. Its design only focuses on obstacle detection in highway scenarios and does not even have clear quantified elevation angular resolution metrics, completely failing to meet the 3D spatial perception requirements in complex scenarios.
More advanced autonomous driving scenarios require the radar to have better elevation resolution. The radar needs to accurately distinguish between drivable areas and structures such as overpasses and footbridges, effectively identify low obstacles on the road surface, accurately recognize curbs across the full FOV, and precisely parse special terrain scenarios such as bridges, slopes, and tunnels, reliably determining open drivable areas. In terms of quantified metrics, the elevation angular resolution is required to be less than 1° at a distance of 100 meters, with stable and reliable height classification capability.
To achieve the above performance goals, the radar needs to expand the vertical perception dimension by increasing the antenna aperture in the vertical direction, adopt advanced beamforming and MIMO technology to increase the number of virtual channels, and introduce high-performance ground reflection suppression technology to eliminate ground clutter interference. It can provide accurate height dimension information for cameras and LiDAR, supporting the stable operation of multi-sensor fusion perception systems.
4. Near-Field Perception
Excellent near-field perception performance is a prerequisite for full-domain environmental perception in dense traffic.
The near-field performance design of traditional radars only serves basic functions such as parking and blind-spot monitoring, with extremely low perception requirements for static environments and very limited lateral coverage. Its minimum reliable detection distance is about 1~2 meters, lacking target discrimination capability at close range, and the forward FOV can only reach 30°~60°, making it difficult to support the refined perception requirements of autonomous driving for the close-range environment around the vehicle.
As the proportion of low-speed urban working conditions in autonomous driving significantly increases, higher demands are placed on the near-field performance of radar. The radar needs to accurately identify static targets immediately adjacent to the ego vehicle, such as curbs and pillars, stably support low-speed driving operations, reliably identify VRUs at close range, and achieve 360° blind-spot-free full-domain coverage, while possessing strong anti-multipath interference capabilities. In terms of quantified metrics, the minimum reliable detection distance is required to be shortened to 0.1~0.2 meters, with clear target discrimination capability within the close range.
To meet the near-field perception requirements, the radar needs targeted optimization of near-field dedicated antenna beams, adopting a design scheme that combines a large FOV with high ranging accuracy and high azimuth angular resolution. It introduces advanced clutter suppression and multipath cancellation technologies, achieves FOV overlapping coverage through dense sensor deployment, and ultimately realizes deep integration of radar, ultrasonic sensors, and cameras to build a complete and reliable close-range perception system.
5. System Latency
Lower system latency can directly enhance the safety, response speed, and driving experience of autonomous driving systems.
The system latency design of traditional radars is only adapted to low-speed dynamic scenarios, with the original design intention mainly to cope with sudden emergencies. All its perception functions are executed locally in the radar ECU (Electronic Control Unit), and the end-to-end latency is mainly sensor internal latency. The corresponding refresh rate is about 5~10 Hz, and the overall latency is usually maintained at 200~300 milliseconds.
In advanced autonomous driving systems, the radar needs to cope with rapidly changing scenarios brought by urban roads and VRUs. Extremely short reaction times become a rigid demand, and system latency thus rises to a key metric affecting perception performance. Meanwhile, multi-sensor fusion requires strict time synchronization of data from each sensor, and the level of latency will directly affect driving safety and comfort. In terms of specific metrics, the refresh rate is required to be increased to 20~40 Hz, and the end-to-end latency target is controlled within 50 milliseconds.
To achieve the above low-latency goals, the radar needs to adopt high-throughput computing units such as hardware accelerators, GPUs (Graphics Processing Units), and DSPs (Digital Signal Processors), build parallel processing pipelines to improve computing efficiency, further compress latency through hardware-software co-design, extract features in advance and carry out early fusion to minimize processing time, and ultimately rely on a high-speed data transmission interconnection architecture to achieve low-latency data interaction.
03.Evolution of Forward-Looking Radar Architecture
1. Driving Factors
The steadily increasing perception complexity of autonomous driving application scenarios has become the core driving force for the continuous iteration of forward-looking radars.
Traditional forward-looking radars are mainly developed around ACC and AEB functions, with high-speed car following as the core application scenario.
Products generally adopt a narrow FOV design, with functions focusing on ranging and velocity measurement capabilities, and relatively low azimuth angular resolution metrics. Meanwhile, perception computing and functional modules are integrated into the radar body itself, generally adopting a local radar processing mode.
The design orientation of new forward-looking radars focuses on three core directions:
Expanding detection coverage: Adopting a large FOV design to simultaneously adapt to high-speed roads and complex urban driving scenarios;
Enhancing spatial perception capabilities: Imposing higher requirements on azimuth angular resolution, elevation angular resolution, dynamic range, and near-field performance;
Optimizing real-time response capabilities: Achieving low latency and high refresh rates, and supporting high-speed data transmission.
2. Development Roadmap of Forward-Looking Radar
The development trend of forward-looking radars in the industry has gradually shifted from simple ranging perception to full-domain spatial environment perception.
It is divided into five generations of product forms:
1) Traditional forward-looking radar: The maximum number of virtual channels is 16, with mainstream specifications of 3T4R and 4T4R.
This type of radar has a balanced cost-to-performance ratio, suitable for large-scale mass production and deployment, but has inherent shortcomings in elevation detection capability. Product design prioritizes ensuring horizontal azimuth measurement while weakening vertical dimension perception performance.
2) Early long-range radar: The maximum number of virtual channels is 64, with mainstream specifications including 6T6R, 4T8R, and 8T8R.
It is positioned as a transitional product for the scaled iteration of radar architecture, and its channel performance still cannot support reliable elevation perception. If the cost advantage can cover the performance shortcomings, this solution still has practical value.
3) First-generation imaging radar: The maximum number of virtual channels is 256, with mainstream specifications of 8T16R, 12T16R, and 16T16R.
The spatial environment perception capability of the product is significantly improved. The increase in the number of virtual channels can be directly translated into perception performance gains, and high-density radar point clouds can meet the parsing requirements of various scenarios.
4) Next-generation imaging radar: The maximum number of virtual channels is 1024, with mainstream specifications of 24T24R and 24T32R.
This product is the preferred solution for high-performance forward-looking radars at this stage. The richness of the radar point cloud is further improved, and under this channel specification, performance upgrades still have investment value.
5) Ultra-Large Array (ULA) imaging radar: The number of virtual channels can reach thousands, with mainstream specifications of 40T40R, 48T48R, and 60T60R.
Relying on ultra-large arrays and low-sidelobe technology, the product can achieve ultra-high spatial fidelity. However, it simultaneously faces systematic challenges such as power consumption, computing power, synchronization, and calibration, and the growth rate of hardware costs far exceeds the performance improvement margin.
3. Channel Count is Not the Only Consideration
The number of virtual channels is not the only metric determining radar performance; modern radar performance improvement relies more on balanced optimization of the entire system. In addition to the channel count, six major design dimensions need to be comprehensively considered:
Among them:
RF and signal design: Covering modulation strategies, pure MIMO operation modes, channel separation capability, and ranging accuracy;
Antenna design: Including antenna layout (sparse array / dense array), FOV shaping, and horizontal and vertical dimension expansion;
Vehicle integration and packaging: Considering the performance impact brought by the radome and bumper, as well as the effects of antenna aperture, installation space, and sensor layout on the FOV;
System constraints: Including power consumption, heat dissipation, anti-interference capability, and calibration scalability;
Frequency bandwidth: Involving bandwidth expansion, high-frequency miniaturization, and multi-band compatible design;
Perception architecture: Including independent perception, high-density point clouds, end-to-end AI perception, scalable fusion architecture, and computing power allocation mechanism.
4. Distributed Radar
Distributed radar possesses unprecedented performance and coverage potential, but its scaled implementation requires overcoming a series of systematic challenges.
Potential advantages include:
Scalability: Adopting modular radar probe design to support flexible expansion of the sensor architecture;
Virtual aperture: Utilizing the front of the vehicle to form an ultra-large virtual aperture, effectively improving spatial resolution capability;
Integration flexibility: Flexible sensor layout methods can adapt to different vehicle design requirements.
Existing challenges:
The system implementation difficulty is very high, with huge engineering challenges in carrier-wave-level phase synchronization, and significantly increased difficulty in wiring, calibration, and verification work;
The layout form of sparse arrays will constrain the upper limit benefits of performance;
Radar performance is easily affected by factors such as installation deviations and equipment aging deformation.
04.Evolution of Surround-View Radar Architecture
1. Driving Factors
The positioning of modern surround-view radar has shifted from an independent functional sensor to a core node in a collaborative perception network.
Traditional corner radar functions focus on blind-spot monitoring, rear collision warning, and rear cross-traffic alert, mainly based on short-range detection. Meanwhile, each radar works independently, with no collaborative capability between sensors.
Modern surround-view radar can achieve:
Full-domain perception: Achieving 360° continuous perception, adaptable to complex scenarios such as urban driving, VRU interaction, and dense traffic;
Collaborative perception: Supporting continuous tracking of targets around the vehicle, achieving trajectory handover between radars and radar-level data fusion;
Collaborative detection: Utilizing FOV overlapping design to reduce perception blind spots while forming sensor redundancy.
Multi-radar collaborative work can build a brand-new environmental perception capability, specifically reflected in three aspects:
Perception capability: Achieving multi-perspective observation to form a unified environmental understanding;
Tracking stability: Supporting 360° full-domain stable tracking, capable of completing cross-radar target cross-validation and achieving trajectory handover according to scenarios;
System response: Shortening response time and improving target detection reliability.
2. Dynamic Virtual Radar Grouping Technology
Dynamic virtual radar grouping technology can fully release the collaborative detection capability of surround-view radar.
Driving scenarios are in a state of dynamic change, and perception priorities are adjusted accordingly. Computing resources are dynamically allocated according to perception needs, and the collaborative detection mode can adaptively switch with scenarios. The system supports the simultaneous operation of multiple virtual radar groups.
Adopting a centralized architecture to coordinate distributed radar resources, the boundaries of physical sensors are weakened at the perception level. The system can achieve unified environmental perception and collaborative tracking, and dynamically adjust waveforms and detection priorities.
However, due to the large number of radars working simultaneously and the large-area overlap of the FOV of each radar, the challenge of interference issues will be greater.
Interference will generate false targets and cause measurement data distortion, leading to a decrease in radar detection robustness and degradation of dynamic perception performance, while simultaneously squeezing the suppression margin for external interference.
At this stage, it mainly relies on centralized overall management, using time-division multiplexing, frequency-division multiplexing, intelligent waveform scheduling, and global interference perception to handle interference.
In the future, optimization solutions such as scenario-adaptive radar working modes, dynamic waveform management, and AI-assisted interference detection and suppression will be further implemented.
05.Centralized Radar Architecture (Satellite Radar)
1. Radar Architecture Solutions
There have been reports of mass production of satellite radar by multiple companies, and more vehicle models will be launched successively recently. Radar systems are gradually evolving from the "sensor-independent intelligence" mode to the "centralized full-domain environmental perception" mode. This trend is becoming increasingly obvious, especially in advanced autonomous driving systems.
The differences between different processing architectures are as follows:
Among them:
"Distributed processing architecture": Each radar is equipped with an independent ECU, perception functions run independently, there is very little interaction between radars, and it directly outputs various data for actuators;
"Transitional processing architecture": Target extraction is completed at the radar end, data is aggregated to the central ECU to achieve target-level fusion, radar collaboration is gradually improved, and a hybrid processing mode is adopted overall;
"Centralized processing architecture": The radar is only responsible for raw data collection, all data is aggregated to the central unit to achieve early data fusion and unified full-domain environmental perception, while completing inter-radar trajectory collaboration, computing power sharing, and system-level scheduling.
The different levels of radar output are shown below, gradually tending towards the raw data form from the upper layer to the lower layer. More raw data means richer radar data representation, and therefore can excavate more detailed information about the environment and targets:
The "distributed architecture" mainly uses actuation signals and tracked targets:
1) Actuation signals: Including HMI (Human-Machine Interface) signals, acceleration signals, and braking signals;
2) Tracked targets: Presented in the form of a target list, including trajectory and classification information, with low bandwidth occupation and limited perception flexibility, only supporting late fusion mode;
The "transitional processing architecture" uses point cloud data: Including point-by-point distance, Doppler, and angle features, with moderate bandwidth occupation, supporting early fusion mode;
The "centralized processing architecture" uses raw radar data: Signals at the ADC (Analog-to-Digital Converter) and FFT (Fast Fourier Transform) levels, existing in the form of sparse 4D tensors; with high bandwidth requirements and the strongest perception flexibility, supporting native AI radar processing and possessing complete secondary reprocessing capabilities.
2. Core Value of Centralized Processing
Centralized processing can integrate multiple radars into a unified environmental perception system.
There are six core functions:
Collaborative radar operation: Unifying modulation strategies, completing cross-radar synchronization and system-level perception scheduling. Traditional distributed radars transmit waveforms independently and operate individually, which is prone to self-interference and timing asynchronization issues; the centralized architecture uniformly manages the transmission parameters and working timing of all radars, avoiding interference at the system level and ensuring perception consistency.
Dynamic resource allocation: Dividing perception priorities based on scenarios to achieve adaptive radar collaboration and flexible perception strategies. The system can dynamically allocate computing power and detection resources according to real-time driving scenarios to maximize computing power utilization.
360° full-domain perception: Achieving continuous target perception and cross-radar trajectory handover to reduce the problem of perception fragmentation. It can solve the defects of trajectory breakage and loss when targets cross radar FOVs in the distributed architecture, complete seamless target handover, and eliminate perception blind spots.
Native AI perception: Equipped with radar point cloud AI algorithms, adopting an occupancy grid-based perception method, and using learnable radar features to complete analysis. AI models are deployed in the central unit, directly processing raw radar point cloud data to improve the accuracy of VRU recognition and drivable area detection.
Early radar fusion: Sharing spatial and Doppler contexts to achieve cross-radar perception and accurately distinguish close-range targets. Different from traditional target-level fusion, this mode completes fusion at the data layer, which can effectively identify weak targets around strong reflectors and reduce the probability of missed and false detections.
Global environmental model: Collaboratively utilizing overlapping FOVs to form rich environmental representation and unified scene understanding. Relying on full-domain data to build environmental models, avoiding conflicts in output information from different radars, and providing stable input for autonomous driving decision-making and control modules.
3. AI Radar Processing Chain
Among them, AI is the most attractive technical direction for the centralized radar architecture.
The traditional radar processing chain is: FFT processing → target detection → preprocessing → point cloud generation → clustering → target tracking → target classification (distinguishing vehicles, pedestrians, cyclists);
The processing chain of AI radar is: Intelligent signal processing (FFTRadNet, ADCNet) → 3D detection (RadDet) → 3D point detection (PointPillars) → scene understanding (PointTransformer, PointNet++) → temporal modeling and target tracking (ConvLSTM, R-CNN).
With the deployment of the centralized radar architecture, traditional radar processing modules are being replaced by neural network architectures.
The native AI end-to-end processing architecture relies on intelligent processing models to achieve early fusion of multiple sensors, making perception results more robust and accurate, while improving algorithm generalization capabilities.
4. Challenges of Centralized Architecture
The centralized architecture also faces many challenges, mainly in the following 3 aspects:
Among them:
Resource and system constraints: Sensor data bandwidth is large, and the scalability of the in-vehicle Ethernet backbone network is insufficient; there are limitations in computing power expansion, heat dissipation, and power consumption, with strict system latency requirements and high difficulty in multi-sensor synchronization;
Functional safety and redundancy: Involving ASIL (Automotive Safety Integrity Level) decomposition, fault degradation, fail-operational architecture, redundant perception links, and fail-safe design, with high overall verification complexity;
Cost and platform adaptation: Need to balance the differentiated deployment of high-end and low-end vehicle models, optimize material costs, achieve functional differentiation, platform architecture, and cross-model software reuse, and achieve cost-efficiency in computing power allocation, that is, optimizing scheduling and reducing hardware costs and power consumption under the premise of meeting performance requirements.
06、Conclusion
Modern radar can achieve multiple types of perception functions, including target detection, scene understanding, temporal tracking, fleet tracking, road curvature estimation, road structure recognition, infrastructure classification, micro-Doppler analysis, dynamic target semantics, scene situation, 3D occupancy grid, free space recognition, ego vehicle motion estimation, static target recognition, landmark mapping, etc., ultimately forming unified full-domain environmental perception.
Looking ahead:
Continuously increasing application value of radar: The number of adapted driving functions continues to increase, the requirements for perception reliability and availability are simultaneously improved, and the number of radars equipped per vehicle gradually grows;
Comprehensive upgrade of performance metrics: Azimuth angular resolution, elevation angular resolution, detection range, and FOV are all upgraded, Doppler information is deeply utilized, and high-density point clouds support refined scene parsing;
System architecture moving towards fusion: Each radar shares a high-performance computing platform, data interoperability is realized between devices, and the degree of sensor fusion is deepened. The technical route has upgraded from raw data processing to native AI architecture and full-domain 4D perception. The evolution of the radar industry is no longer driven solely by single-sensor performance; scalability under the vehicle architecture has become a core consideration.
Note: All images in the article are from "What's Driving the Evolution of Automotive Radar Architectures? System-Level Drivers for ADAS and Automated Driving Saad Nawaz | Shanghai | May 29, 2026". Some images have been appropriately adjusted to improve visualization. Some viewpoints in the original report may not be fully covered; other supplementary information can be found in the extended reading below.
Personal opinions, not necessarily accurate, discussions are welcome.
I am Xueling, researching the technology, products, and applications of perception, control, and AI. Discussions are welcome. For contacting Xueling and the full collection index, please refer to: https://dcn7get8fskg.feishu.cn/wiki/CCMpwjC0EiBIw2kFr7uc84qHneb