Request Access
Technical White Paper: Design and Implementation of a Tele-Operated Data Acquisition Platform for Autonomous Localization in Arid Environments
Technical White Paper: Design and Implementation of a Tele-Operated Data Acquisition Platform for Autonomous Localization in Arid Environments
Author: Ramez Alghazawi, B.Eng.
Organization: ReDrive Systems
Date: November 2025
Abstract
The rapid deployment of Autonomous Vehicles (AVs) is currently hindered by the "domain gap"—the discrepancy between training data collected in temperate zones (Europe/USA) and the environmental realities of arid regions like the UAE. This paper outlines the technical architecture for ReDrive Systems’ initial data collection and teleoperation vehicle. We define the selection of the Hyundai Ioniq 5 platform, the integration of a multi-modal sensor suite (LiDAR, Radar, Camera) via the NVIDIA Jetson compute architecture, and a low-latency 5G teleoperation pipeline. The proposed system aims to generate a localized 50,000 km dataset while validating remote intervention capabilities. The initial pilot phase allocates 6 months for hardware integration and calibration, followed by 6 months of intensive data acquisition in high-temperature and particulate-heavy (sand) conditions.
1. Vehicle Platform Selection & Justification
1.1 The Base Platform: Hyundai Ioniq 5
The project utilizes the Hyundai Ioniq 5 as the core research platform. This selection is supported by three primary engineering factors:
E-GMP Architecture (High Voltage Stability): Unlike Internal Combustion Engine (ICE) vehicles which rely on alternators with fluctuating voltage, the Ioniq 5’s 77.4 kWh battery pack provides a stable high-voltage DC source. This is critical for powering high-load sensors (LiDAR) and compute units without requiring a secondary generator [1].
Drive-by-Wire (DbW) Compatibility: The Ioniq 5 features modern EPS (Electronic Power Steering) and electronic braking systems that are readily accessible via CAN (Controller Area Network) bus manipulation. This allows for the integration of Dataspeed or PolySync DbW kits, enabling the teleoperation station to control steering, throttle, and brake digitally with <10ms actuation latency.
Thermal Management: The vehicle's advanced liquid cooling system (intended for the battery) can be tapped or replicated to cool the trunk-mounted compute stack, which is vital in UAE ambient temperatures exceeding 45°C.
2. Hardware Architecture & Sensor Fusion
To capture the "Ground Truth" needed for UAE-specific training, the vehicle requires a sensor suite that mimics Level 4 autonomy.
2.1 Compute Unit: NVIDIA Jetson AGX Orin
Role: Sensor Fusion, Data Logging, and Video Encoding (H.265) for Teleoperation.
Justification: The AGX Orin provides 275 TOPS (Trillions of Operations Per Second). It supports the high bandwidth required to ingest 8+ camera streams and 3D LiDAR point clouds simultaneously while maintaining a small thermal and energy footprint compared to x86 server racks [2].
2.2 Sensor Suite Specification
Based on the €150,000 integration budget, the following configuration offers the highest data fidelity:
LiDAR (Light Detection and Ranging):
Hardware: Ouster OS1-128 (Rev 7) or Hesai XT32.
Scientific Function: 360° mapping. The Ouster OS1 is specifically chosen for its IP69K rating and ability to operate in high heat, unlike older mechanical spinners which fail in dust/sand.
Cameras (Visual Spectrum):
Hardware: 6x GMSL2 (Gigabit Multimedia Serial Link) Cameras (e.g., Leopard Imaging).
Placement: Front (Wide/Narrow), Sides, Rear.
Protocol: GMSL2 is mandatory over USB cameras to ensure synchronized global shuttering and sub-millisecond latency.
Radar:
Hardware: Continental ARS 408 (77GHz).
Function: Long-range object detection that remains robust during sandstorms where LiDAR wavelengths may scatter [3].
Author: Ramez Alghazawi, B.Eng.
Organization: ReDrive Systems
Date: November 2025
Abstract
The rapid deployment of Autonomous Vehicles (AVs) is currently hindered by the "domain gap"—the discrepancy between training data collected in temperate zones (Europe/USA) and the environmental realities of arid regions like the UAE. This paper outlines the technical architecture for ReDrive Systems’ initial data collection and teleoperation vehicle. We define the selection of the Hyundai Ioniq 5 platform, the integration of a multi-modal sensor suite (LiDAR, Radar, Camera) via the NVIDIA Jetson compute architecture, and a low-latency 5G teleoperation pipeline. The proposed system aims to generate a localized 50,000 km dataset while validating remote intervention capabilities. The initial pilot phase allocates 6 months for hardware integration and calibration, followed by 6 months of intensive data acquisition in high-temperature and particulate-heavy (sand) conditions.
1. Vehicle Platform Selection & Justification
1.1 The Base Platform: Hyundai Ioniq 5
The project utilizes the Hyundai Ioniq 5 as the core research platform. This selection is supported by three primary engineering factors:
E-GMP Architecture (High Voltage Stability): Unlike Internal Combustion Engine (ICE) vehicles which rely on alternators with fluctuating voltage, the Ioniq 5’s 77.4 kWh battery pack provides a stable high-voltage DC source. This is critical for powering high-load sensors (LiDAR) and compute units without requiring a secondary generator [1].
Drive-by-Wire (DbW) Compatibility: The Ioniq 5 features modern EPS (Electronic Power Steering) and electronic braking systems that are readily accessible via CAN (Controller Area Network) bus manipulation. This allows for the integration of Dataspeed or PolySync DbW kits, enabling the teleoperation station to control steering, throttle, and brake digitally with <10ms actuation latency.
Thermal Management: The vehicle's advanced liquid cooling system (intended for the battery) can be tapped or replicated to cool the trunk-mounted compute stack, which is vital in UAE ambient temperatures exceeding 45°C.
2. Hardware Architecture & Sensor Fusion
To capture the "Ground Truth" needed for UAE-specific training, the vehicle requires a sensor suite that mimics Level 4 autonomy.
2.1 Compute Unit: NVIDIA Jetson AGX Orin
Role: Sensor Fusion, Data Logging, and Video Encoding (H.265) for Teleoperation.
Justification: The AGX Orin provides 275 TOPS (Trillions of Operations Per Second). It supports the high bandwidth required to ingest 8+ camera streams and 3D LiDAR point clouds simultaneously while maintaining a small thermal and energy footprint compared to x86 server racks [2].
2.2 Sensor Suite Specification
Based on the €150,000 integration budget, the following configuration offers the highest data fidelity:
LiDAR (Light Detection and Ranging):
Hardware: Ouster OS1-128 (Rev 7) or Hesai XT32.
Scientific Function: 360° mapping. The Ouster OS1 is specifically chosen for its IP69K rating and ability to operate in high heat, unlike older mechanical spinners which fail in dust/sand.
Cameras (Visual Spectrum):
Hardware: 6x GMSL2 (Gigabit Multimedia Serial Link) Cameras (e.g., Leopard Imaging).
Placement: Front (Wide/Narrow), Sides, Rear.
Protocol: GMSL2 is mandatory over USB cameras to ensure synchronized global shuttering and sub-millisecond latency.
Radar:
Hardware: Continental ARS 408 (77GHz).
Function: Long-range object detection that remains robust during sandstorms where LiDAR wavelengths may scatter [3].
Author: Ramez Alghazawi, B.Eng.
Organization: ReDrive Systems
Date: November 2025
Abstract
The rapid deployment of Autonomous Vehicles (AVs) is currently hindered by the "domain gap"—the discrepancy between training data collected in temperate zones (Europe/USA) and the environmental realities of arid regions like the UAE. This paper outlines the technical architecture for ReDrive Systems’ initial data collection and teleoperation vehicle. We define the selection of the Hyundai Ioniq 5 platform, the integration of a multi-modal sensor suite (LiDAR, Radar, Camera) via the NVIDIA Jetson compute architecture, and a low-latency 5G teleoperation pipeline. The proposed system aims to generate a localized 50,000 km dataset while validating remote intervention capabilities. The initial pilot phase allocates 6 months for hardware integration and calibration, followed by 6 months of intensive data acquisition in high-temperature and particulate-heavy (sand) conditions.
1. Vehicle Platform Selection & Justification
1.1 The Base Platform: Hyundai Ioniq 5
The project utilizes the Hyundai Ioniq 5 as the core research platform. This selection is supported by three primary engineering factors:
E-GMP Architecture (High Voltage Stability): Unlike Internal Combustion Engine (ICE) vehicles which rely on alternators with fluctuating voltage, the Ioniq 5’s 77.4 kWh battery pack provides a stable high-voltage DC source. This is critical for powering high-load sensors (LiDAR) and compute units without requiring a secondary generator [1].
Drive-by-Wire (DbW) Compatibility: The Ioniq 5 features modern EPS (Electronic Power Steering) and electronic braking systems that are readily accessible via CAN (Controller Area Network) bus manipulation. This allows for the integration of Dataspeed or PolySync DbW kits, enabling the teleoperation station to control steering, throttle, and brake digitally with <10ms actuation latency.
Thermal Management: The vehicle's advanced liquid cooling system (intended for the battery) can be tapped or replicated to cool the trunk-mounted compute stack, which is vital in UAE ambient temperatures exceeding 45°C.
2. Hardware Architecture & Sensor Fusion
To capture the "Ground Truth" needed for UAE-specific training, the vehicle requires a sensor suite that mimics Level 4 autonomy.
2.1 Compute Unit: NVIDIA Jetson AGX Orin
Role: Sensor Fusion, Data Logging, and Video Encoding (H.265) for Teleoperation.
Justification: The AGX Orin provides 275 TOPS (Trillions of Operations Per Second). It supports the high bandwidth required to ingest 8+ camera streams and 3D LiDAR point clouds simultaneously while maintaining a small thermal and energy footprint compared to x86 server racks [2].
2.2 Sensor Suite Specification
Based on the €150,000 integration budget, the following configuration offers the highest data fidelity:
LiDAR (Light Detection and Ranging):
Hardware: Ouster OS1-128 (Rev 7) or Hesai XT32.
Scientific Function: 360° mapping. The Ouster OS1 is specifically chosen for its IP69K rating and ability to operate in high heat, unlike older mechanical spinners which fail in dust/sand.
Cameras (Visual Spectrum):
Hardware: 6x GMSL2 (Gigabit Multimedia Serial Link) Cameras (e.g., Leopard Imaging).
Placement: Front (Wide/Narrow), Sides, Rear.
Protocol: GMSL2 is mandatory over USB cameras to ensure synchronized global shuttering and sub-millisecond latency.
Radar:
Hardware: Continental ARS 408 (77GHz).
Function: Long-range object detection that remains robust during sandstorms where LiDAR wavelengths may scatter [3].
[Diagram 1 : Vehicle Topology]
(Description): A block diagram showing the Hyundai Ioniq 5 Battery connecting to a DC-DC Converter (stepping down to 12V/24V). This powers the Jetson AGX Orin. The Jetson is connected via Ethernet to the LiDAR and via FAKRA cables to 6 Cameras. The CAN Bus Interface connects the Jetson to the car's steering/brakes.
3. Teleoperation & Connectivity Architecture
The safety of the system relies on the "Human-in-the-Loop" via a remote driving station. The primary engineering challenge is Glass-to-Glass Latency (the time from the camera capturing an image to the operator seeing it).
3.1 The Connectivity Link (5G Bonding)
A single SIM card is insufficient for safety-critical video streaming.
Hardware: Peplink MAX BR1 Pro 5G or similar Multi-WAN Router.
Methodology: "WAN Bonding." We utilize multiple SIM cards from different UAE carriers (e.g., Etisalat and Du) simultaneously. The router splits the video packets across all connections. If one carrier drops, the video stream continues uninterrupted.
Target Latency: <100ms round-trip time (RTT).
3.2 The Tele-Driving Station
Located at the ReDrive control center (budgeted at €10,000).
Input Hardware: Logitech G29 or Fanatec steering wheel with force feedback.
Software Stack: The station runs a decoding client (WebRTC based) that receives the video stream and sends control commands (Steering Angle, Torque, Brake Pressure) back to the vehicle via MQTT or UDP protocols.
[Diagram 1 : Vehicle Topology]
(Description): A block diagram showing the Hyundai Ioniq 5 Battery connecting to a DC-DC Converter (stepping down to 12V/24V). This powers the Jetson AGX Orin. The Jetson is connected via Ethernet to the LiDAR and via FAKRA cables to 6 Cameras. The CAN Bus Interface connects the Jetson to the car's steering/brakes.
3. Teleoperation & Connectivity Architecture
The safety of the system relies on the "Human-in-the-Loop" via a remote driving station. The primary engineering challenge is Glass-to-Glass Latency (the time from the camera capturing an image to the operator seeing it).
3.1 The Connectivity Link (5G Bonding)
A single SIM card is insufficient for safety-critical video streaming.
Hardware: Peplink MAX BR1 Pro 5G or similar Multi-WAN Router.
Methodology: "WAN Bonding." We utilize multiple SIM cards from different UAE carriers (e.g., Etisalat and Du) simultaneously. The router splits the video packets across all connections. If one carrier drops, the video stream continues uninterrupted.
Target Latency: <100ms round-trip time (RTT).
3.2 The Tele-Driving Station
Located at the ReDrive control center (budgeted at €10,000).
Input Hardware: Logitech G29 or Fanatec steering wheel with force feedback.
Software Stack: The station runs a decoding client (WebRTC based) that receives the video stream and sends control commands (Steering Angle, Torque, Brake Pressure) back to the vehicle via MQTT or UDP protocols.
[Diagram 1 : Vehicle Topology]
(Description): A block diagram showing the Hyundai Ioniq 5 Battery connecting to a DC-DC Converter (stepping down to 12V/24V). This powers the Jetson AGX Orin. The Jetson is connected via Ethernet to the LiDAR and via FAKRA cables to 6 Cameras. The CAN Bus Interface connects the Jetson to the car's steering/brakes.
3. Teleoperation & Connectivity Architecture
The safety of the system relies on the "Human-in-the-Loop" via a remote driving station. The primary engineering challenge is Glass-to-Glass Latency (the time from the camera capturing an image to the operator seeing it).
3.1 The Connectivity Link (5G Bonding)
A single SIM card is insufficient for safety-critical video streaming.
Hardware: Peplink MAX BR1 Pro 5G or similar Multi-WAN Router.
Methodology: "WAN Bonding." We utilize multiple SIM cards from different UAE carriers (e.g., Etisalat and Du) simultaneously. The router splits the video packets across all connections. If one carrier drops, the video stream continues uninterrupted.
Target Latency: <100ms round-trip time (RTT).
3.2 The Tele-Driving Station
Located at the ReDrive control center (budgeted at €10,000).
Input Hardware: Logitech G29 or Fanatec steering wheel with force feedback.
Software Stack: The station runs a decoding client (WebRTC based) that receives the video stream and sends control commands (Steering Angle, Torque, Brake Pressure) back to the vehicle via MQTT or UDP protocols.
[Diagram 2 : Network Architecture]
(Description):
Vehicle Side: Cameras -> Jetson (Video Encoder) -> 5G Modem -> Cellular Tower.
Cloud/Network: VPN Tunnel / bonding server.
Station Side: Internet -> Decoding PC -> Monitors -> Operator.
Feedback Loop: Operator Steering -> Control Signal -> Internet -> Vehicle -> CAN Bus -> Wheels turn.
4. Cost Feasibility & Budget Alignment
The following breakdown confirms the technical feasibility within the €255,000 funding request (specifically the €150k vehicle portion and €10k station).
Component Specification Est. Cost (€) Justification Base Vehicle Hyundai Ioniq 5 (Used/New) €55,000 High voltage EV, E-GMP platform. Drive-by-Wire Kit Dataspeed or similar €25,000 Necessary for computer control of steering. Compute NVIDIA Jetson AGX Orin Dev Kit + Storage €4,000 Industry standard for edge AI. LiDAR 1x Ouster OS1 (128 channel) €18,000 High resolution for mapping. Visual/Radar 6x Cameras + 1x Radar + Cabling €12,000 Surround perception. Integration Custom mounting, cooling, power dist. €36,000 Labor and fabrication (aluminum extrusion). TOTAL €150,000 Matches Phase 1 Budget
5. Scientific Validity & Environmental Considerations
5.1 The "Sand" Problem (Occlusion)
Standard AV stacks use LiDAR intensity returns to detect lane lines. In the UAE, sand accumulation on roads lowers the retro-reflectivity of markings.
ReDrive Approach: Our dataset will specifically tag "sand-occluded" road markings to train Deep Neural Networks (DNNs) to infer lane geometry from context, rather than just markings. This is a novel contribution to the field of Domain Adaptation [4].
5.2 Thermal Throttling
Compute hardware performance degrades above 80°C junction temperature.
Mitigation: The vehicle build includes an active thermoelectric cooling (Peltier) enclosure for the Jetson unit, ensuring consistent clock speeds during UAE summer operations.
6. Conclusion
The proposed architecture utilizes Commercial Off-The-Shelf (COTS) hardware to build a robust research platform. By leveraging the Hyundai Ioniq 5’s electrical architecture and NVIDIA’s edge computing, ReDrive Systems can successfully capture high-fidelity, synchronized sensor data. The 5G bonded teleoperation layer adds a critical safety redundancy, allowing for immediate deployment on public roads. This system is not merely a vehicle; it is a mobile laboratory capable of solving the specific domain adaptation challenges of the MENA region.
References
G. Massaro et al., "Energy Management Strategies for Autonomous Electric Vehicles," IEEE Transactions on Transportation Electrification, 2021. (Validates EV choice for power stability).
NVIDIA Corporation, "Jetson AGX Orin Series Technical Specifications," 2023. (Validates compute power).
S. Hasarli et al., "Effects of Dust and Sand on LiDAR and Radar Sensors," SAE International Journal of Connected and Automated Vehicles, 2020. (Validates need for multi-modal sensors).
M. Wang, "Unsupervised Domain Adaptation for Semantic Segmentation in Autonomous Driving," CVPR, 2020. (Validates the scientific need for local data).
[Diagram 2 : Network Architecture]
(Description):
Vehicle Side: Cameras -> Jetson (Video Encoder) -> 5G Modem -> Cellular Tower.
Cloud/Network: VPN Tunnel / bonding server.
Station Side: Internet -> Decoding PC -> Monitors -> Operator.
Feedback Loop: Operator Steering -> Control Signal -> Internet -> Vehicle -> CAN Bus -> Wheels turn.
4. Cost Feasibility & Budget Alignment
The following breakdown confirms the technical feasibility within the €255,000 funding request (specifically the €150k vehicle portion and €10k station).
Component Specification Est. Cost (€) Justification Base Vehicle Hyundai Ioniq 5 (Used/New) €55,000 High voltage EV, E-GMP platform. Drive-by-Wire Kit Dataspeed or similar €25,000 Necessary for computer control of steering. Compute NVIDIA Jetson AGX Orin Dev Kit + Storage €4,000 Industry standard for edge AI. LiDAR 1x Ouster OS1 (128 channel) €18,000 High resolution for mapping. Visual/Radar 6x Cameras + 1x Radar + Cabling €12,000 Surround perception. Integration Custom mounting, cooling, power dist. €36,000 Labor and fabrication (aluminum extrusion). TOTAL €150,000 Matches Phase 1 Budget
5. Scientific Validity & Environmental Considerations
5.1 The "Sand" Problem (Occlusion)
Standard AV stacks use LiDAR intensity returns to detect lane lines. In the UAE, sand accumulation on roads lowers the retro-reflectivity of markings.
ReDrive Approach: Our dataset will specifically tag "sand-occluded" road markings to train Deep Neural Networks (DNNs) to infer lane geometry from context, rather than just markings. This is a novel contribution to the field of Domain Adaptation [4].
5.2 Thermal Throttling
Compute hardware performance degrades above 80°C junction temperature.
Mitigation: The vehicle build includes an active thermoelectric cooling (Peltier) enclosure for the Jetson unit, ensuring consistent clock speeds during UAE summer operations.
6. Conclusion
The proposed architecture utilizes Commercial Off-The-Shelf (COTS) hardware to build a robust research platform. By leveraging the Hyundai Ioniq 5’s electrical architecture and NVIDIA’s edge computing, ReDrive Systems can successfully capture high-fidelity, synchronized sensor data. The 5G bonded teleoperation layer adds a critical safety redundancy, allowing for immediate deployment on public roads. This system is not merely a vehicle; it is a mobile laboratory capable of solving the specific domain adaptation challenges of the MENA region.
References
G. Massaro et al., "Energy Management Strategies for Autonomous Electric Vehicles," IEEE Transactions on Transportation Electrification, 2021. (Validates EV choice for power stability).
NVIDIA Corporation, "Jetson AGX Orin Series Technical Specifications," 2023. (Validates compute power).
S. Hasarli et al., "Effects of Dust and Sand on LiDAR and Radar Sensors," SAE International Journal of Connected and Automated Vehicles, 2020. (Validates need for multi-modal sensors).
M. Wang, "Unsupervised Domain Adaptation for Semantic Segmentation in Autonomous Driving," CVPR, 2020. (Validates the scientific need for local data).
[Diagram 2 : Network Architecture]
(Description):
Vehicle Side: Cameras -> Jetson (Video Encoder) -> 5G Modem -> Cellular Tower.
Cloud/Network: VPN Tunnel / bonding server.
Station Side: Internet -> Decoding PC -> Monitors -> Operator.
Feedback Loop: Operator Steering -> Control Signal -> Internet -> Vehicle -> CAN Bus -> Wheels turn.
4. Cost Feasibility & Budget Alignment
The following breakdown confirms the technical feasibility within the €255,000 funding request (specifically the €150k vehicle portion and €10k station).
Component Specification Est. Cost (€) Justification Base Vehicle Hyundai Ioniq 5 (Used/New) €55,000 High voltage EV, E-GMP platform. Drive-by-Wire Kit Dataspeed or similar €25,000 Necessary for computer control of steering. Compute NVIDIA Jetson AGX Orin Dev Kit + Storage €4,000 Industry standard for edge AI. LiDAR 1x Ouster OS1 (128 channel) €18,000 High resolution for mapping. Visual/Radar 6x Cameras + 1x Radar + Cabling €12,000 Surround perception. Integration Custom mounting, cooling, power dist. €36,000 Labor and fabrication (aluminum extrusion). TOTAL €150,000 Matches Phase 1 Budget
5. Scientific Validity & Environmental Considerations
5.1 The "Sand" Problem (Occlusion)
Standard AV stacks use LiDAR intensity returns to detect lane lines. In the UAE, sand accumulation on roads lowers the retro-reflectivity of markings.
ReDrive Approach: Our dataset will specifically tag "sand-occluded" road markings to train Deep Neural Networks (DNNs) to infer lane geometry from context, rather than just markings. This is a novel contribution to the field of Domain Adaptation [4].
5.2 Thermal Throttling
Compute hardware performance degrades above 80°C junction temperature.
Mitigation: The vehicle build includes an active thermoelectric cooling (Peltier) enclosure for the Jetson unit, ensuring consistent clock speeds during UAE summer operations.
6. Conclusion
The proposed architecture utilizes Commercial Off-The-Shelf (COTS) hardware to build a robust research platform. By leveraging the Hyundai Ioniq 5’s electrical architecture and NVIDIA’s edge computing, ReDrive Systems can successfully capture high-fidelity, synchronized sensor data. The 5G bonded teleoperation layer adds a critical safety redundancy, allowing for immediate deployment on public roads. This system is not merely a vehicle; it is a mobile laboratory capable of solving the specific domain adaptation challenges of the MENA region.
References
G. Massaro et al., "Energy Management Strategies for Autonomous Electric Vehicles," IEEE Transactions on Transportation Electrification, 2021. (Validates EV choice for power stability).
NVIDIA Corporation, "Jetson AGX Orin Series Technical Specifications," 2023. (Validates compute power).
S. Hasarli et al., "Effects of Dust and Sand on LiDAR and Radar Sensors," SAE International Journal of Connected and Automated Vehicles, 2020. (Validates need for multi-modal sensors).
M. Wang, "Unsupervised Domain Adaptation for Semantic Segmentation in Autonomous Driving," CVPR, 2020. (Validates the scientific need for local data).