Torc Robotics has provided new details about AV 3.0, the software framework behind its TorcDrive autonomous driving platform, which is designed for SAE Level 4 long-haul trucking.
The company describes AV 3.0 as a “glass box” architecture that combines artificial intelligence with rule-based safety controls, aiming to balance transparency, traceability, and performance in autonomous freight operations.
Combining AI and Rule-Based Systems
According to Torc, AV 3.0 is built around three primary software modules: perception, prediction, and planning.
While these components use AI-driven models, they operate within deterministic guardrails and predefined safety criteria designed to improve transparency and simplify validation.
The company says the architecture allows engineers to inspect intermediate outputs from each module and evaluate system behavior more easily than fully end-to-end autonomous driving models.
Torc argues that the approach combines the interpretability of traditional rule-based systems with the performance benefits of modern AI.
Evolution of Autonomous Driving Architectures
Torc outlined the progression of autonomous driving technology through several generations.
Early autonomous systems relied primarily on rule-based programming, where vehicle decisions could be directly linked to specific instructions and sensor inputs.
The industry later moved toward hybrid systems that combined machine-learning-based perception with rule-based planning and prediction.
More recently, end-to-end AI architectures have emerged, integrating perception, prediction, planning, and control into a single trained model.
While these systems can deliver strong performance, Torc says their internal decision-making processes can be difficult to interpret, creating challenges for debugging, validation, and safety certification.
Designed for Heavy-Duty Trucking
Torc believes autonomous trucking presents unique challenges compared with passenger vehicles.
Heavy trucks require longer stopping distances, have limited maneuverability, and must perceive hazards at greater distances to operate safely.
The company says these operational demands influenced the development of AV 3.0 and its modular design philosophy.
The platform is intended to support SAE Level 4 autonomous operation, where the vehicle can perform all driving functions within a defined operational domain without human intervention.
Extensive Simulation-Based Training
Torc said TorcDrive is trained using a combination of real-world driving data and large-scale simulation.
The company’s simulation platform generates a wide variety of driving scenarios, including interactions with passenger vehicles, commercial trucks, pedestrians, animals, construction zones, and other rare events.
According to Torc, the simulation environment is designed to expose the system to situations that would be difficult or impractical to encounter frequently through real-world testing alone.
The company emphasized that simulation supplements, rather than replaces, on-road validation.
Built With Daimler Truck
Torc became an independent subsidiary of Daimler Truck North America in 2019, establishing a partnership focused on autonomous freight transportation.
The companies have jointly developed a production-intent autonomous truck based on the Freightliner Cascadia platform.
The vehicle combines TorcDrive software, an NVIDIA-powered embedded computing system supplied by Flex, and Daimler’s autonomous-ready Freightliner Cascadia 5.0 architecture.
According to Torc, sensors and safety-critical redundancy systems are integrated during vehicle production rather than added through aftermarket retrofitting.
Path Toward Commercial Autonomous Freight
Torc positions AV 3.0 as a foundation for future commercial deployment of autonomous trucking services.
The company says its “glass box” approach is intended to provide the performance benefits of advanced AI while maintaining the transparency required for validation, safety assurance, and regulatory acceptance.
As autonomous trucking developers move closer to commercial operations, explainability and safety verification are becoming increasingly important factors alongside driving performance and operational efficiency.