The new L234 vehicle platform from General Motors (GM), on which the Cadillac Vistiq is based, is equipped with a Level 2+ ADAS system. Eleven cameras, one radar, and one lidar sensor are used for environmental perception. The driving functions (motion control) use this sensor data to control electric actuators such as the electronic brake and electric power steering (EPS). The goal of GM’s validation project is the robust verification of all Level 2+ functions under deterministic real-time conditions and taking the specific requirements of regional markets into consideration.

How can automated driving functions be robustly and systematically validated?

Autonomous and highly automated driving functions require full validation of the entire functional chain – from sensors through perception and decision logic to actuators.

GM achieves this through:

•  Established safety-oriented development processes
• Simulation-based functional evaluation 
• Hardware-in-the-loop (HIL) tests with highest accuracy and real-time conditions
• The goal is a reproducible, standards-compliant release at the ECU and system levels.

How is cross-domain HIL validation achieved?

For cross-domain validation, GM uses a modular HIL architecture consisting of:

• dSPACE HIL simulators for real-time model execution and ECU integration
• Mechatronic test benches (e.g., brake and steering systems) for coupling real actuators into the test setup
• Dedicated gateways for fault injection at both signal and vehicle network levels

This modular architecture enables deterministic and repeatable test execution, as well as high-resolution analysis of functional behavior and safety mechanisms across multiple vehicle domains.

The integration of mechatronic test benches is essential to extend classical ECU-level HIL testing toward physical closed-loop validation. By coupling real brake and steering actuators into the HIL setup, GM is able to validate not only control logic and communication timing, but also actuator dynamics, physical load behavior, and safety-critical interactions between software and hardware under realistic operating conditions.
 

How can dynamic traffic and environment scenarios be generated?

Realistic traffic and environment scenarios are essential for the validation of ADAS functions. For this purpose, the simulation tools from dSPACE play a key role in validation at GM. The Automotive Simulation Models (ASM) tool suite is used to simulate a wide range of traffic scenarios, including relevant edge cases. Traffic scenarios are drawn from a predefined scenario library and can be automatically varied through parameterization. Based on the traffic simulation, AURELION generates a sensor-realistic, fully immersive 3D environment that produces physically accurate sensor data such as camera raw data and lidar point cloud data.

Together, ASM and AURELION enable:

• Physically correct vehicle dynamics behavior and maneuvers for ego-vehicles and traffic vehicles
• High variability and scalability of scenarios to increase test coverage
• Sensor-realistic environment and sensor frontend simulation with validated camera, radar, and lidar models, (as well as radar simulation based on object lists via CAN)
• Deterministic generation of camera raw data and lidar point clouds, including complex and safety-critical traffic situations

This integrated approach allows the consistent, repeatable, and reproducible creation of complex dynamic traffic and environment scenarios for reliable ADAS validation in a HIL environment.
 

How can ADAS sensors be integrated into HIL simulation?

Depending on the test objective, several sensor-integration options are available in HIL simulation. The appropriate approach is determined by the required level of realism and by which part of the sensor system is to be validated:

• For maximum realism, over-the-air (OTA) stimulation of the physical sensor can be used.
• A highly realistic alternative is sensor simulation at raw data level fed into the corresponding ECU inputs.
• For many validation tasks, ideal ground-truth sensor simulation is sufficient, where sensor outputs (e.g., object lists) are simulated and provided directly to the ADAS ECU.

In GM’s project setup, camera sensors play a particularly important role. Therefore, a high degree of realism in the sensor raw data was required. Raw-data-based sensor simulation was selected to enable realistic and vehicle-representative perception validation.

Integration via raw-data injection:

In the case of raw-data injection, ADAS sensors impose specific integration requirements due to vendor-specific interfaces, protocols, and sensor metadata. The dSPACE tool chain addresses these challenges through a continuous sensor processing chain: Sensor raw data generated in AURELION is processed in the Environment Sensor Interface (ESI) Unit (Environment Sensor Interface Unit), adapted as required, and transmitted to the ADAS ECU via the appropriate physical interface and protocol. Camera signals are transferred via ESI Units to the ADAS ECUs, enabling perception-level validation under vehicle-representative conditions.

The following sensor-specific properties are taken into account:

• Timing and synchronization signals
• Sensor metadata, such as exposure control and histogram information
• Standardized SerDes (Serializer/Deserializer) interfaces and protocols, e.g., MIPI A-PHY and ASA-ML
• Proprietary SerDes interfaces and protocols, e.g., ADI GMSL and TI FPD-Link

Using ESI Units, GM validated the following sensor configurations in the HIL environment:

• 4 × 2 MP (megapixel) fisheye cameras
• 5 × 3 MP cameras
• 2 × 8 MP cameras

The camera model could be parameterized for the individual camera types based on their data sheets, enabling accurate and type-specific camera simulation.

The ESI Units ensure bit-accurate and timing-accurate sensor simulation, enabling faithful integration of the camera sensors into the closed-loop HIL test setup.

Integration at the network level:

• The lidar point clouds simulated with AURELION are transferred to the ECU via Ethernet.
• Radar sensors were integrated using simulated radar targets, generated as object lists and transmitted to the ECU via CAN.

This synchronized approach enables traffic and environment simulation for ADAS that fulfills all plausibility requirements and ensures that the ADAS ECU operates under the same conditions as in the vehicle. As a result, GM was able to validate perception and downstream control logic under vehicle-representative sensor conditions while maintaining full determinism and reproducibility.
 

Insights into the ADAS and chassis HIL architecture

The integrated platform for HIL testing spans both ADAS and chassis domains and covers:

ADAS ECUs and sensors

• Vehicle dynamics and traffic simulation using the ASM (Automotive Simulation Models) tool suite
• Validated environment and sensor models, implemented in
  • AURELION (GPU-based) for high-fidelity sensor and environment simulation
  • ESI Unit (FPGA-based) for deterministic, low-latency sensor signal injection 
• Real-time validation of perception, fusion, and decision logic using the exact same communication interfaces and protocols (FPD-Link, Ethernet), as in the vehicle

Electronic Brake Control Module (EBCM including ABS, TCS, ESP)

• Mechatronic bench provided by GM
• Simulation of wheel speed, longitudinal and lateral acceleration, and yaw rate
• Evaluation of hydraulic-electric actuator behavior under closed-loop conditions

Electric Power Steering (EPS)

• Mechatronic bench provided by dSPACE
• Road and chassis models for realistic steering torque simulation
• Validation of safety-relevant control logic, including fallback strategies and torque limitation functions

Network & Diagnostics

• Full integration of in-vehicle networks, including CAN and Ethernet 
• Restbus simulation for all relevant communication nodes 
• Diagnostic and communication analysis under functional and fault conditions