At Torc Robotics, we’re at the forefront of self-driving truck technology. Our pursuit of innovation is underpinned by a comprehensive validation strategy that seeks to prove the feasibility of our self-driving truck product. Today, we’re diving into our validation approach, exploring the various forms of proof we employ, the criteria for achieving true Level 4 readiness, and the multi-pronged validation strategy that drives our groundbreaking work.
Exploring the Self-Driving Challenge
Our validation strategy is supported by three core pillars: problem definition, current references, and proof.
Understanding the Problem
At the heart of Torc’s validation strategy is a clear definition of the self-driving challenge we’re addressing. By precisely outlining the complexities and intricacies of self-driving trucks, we lay the groundwork for our validation efforts.
Understanding the problem begins with problem completeness. The operating domain is defined prior, with manageable parameters and modellable relationships. IFTDs, or In-Vehicle Fallback Test Drivers, provide source data of an ideal truck driver, allowing us to provide driving behaviors that correlate with a non-robotic driver’s ability.
Our on-the-field teams act as a solid reference model for many aspects of our self-driving technology, including our validation strategy.
Reference Models
We rely on a number of reference models to understand the whole problem, including In-Vehicle Fallback Test Drivers (IFTDs), laws, voice of the customer, and more.
In the case of our IFTDs, these professionals act as an integral piece of our validation process. These highly trained individuals are CDL-holding drivers with years of experience driving for logistics leaders across the United States; their driving behaviors are ideal resources for robotic truck behavior, giving us an effective reference point throughout software development.
Proof: Rigorous Testing and Pushing Boundaries
Our commitment to creating a safe, scalable self-driving truck extends beyond confirming functionality; we deliberately attempt to break our technology to reveal potential vulnerabilities. We employ various forms of proof:
- Direct Proof Based on Requirements. Data collected from test runs with our in-house semi-trucks forms the basis for formal testing. This includes techniques like black box testing and ad-hoc testing to comprehensively address anticipated challenges.
- Proof by Exhaustion. We subject our system to an exhaustive range of scenarios, leveraging simulations to expand testing without resource constraints.
- Proof by Contradiction. We intentionally introduce incorrect data to test the system’s adaptability. For instance, we might challenge the system with non-moving objects mimicking high-speed movement, feed two sensors different datasets, or otherwise attempt to “confuse” the autonomous driving system.
- Proof by Random. Our technology’s versatility is tested by placing it in unfamiliar environments, evaluating its ability to handle unforeseen scenarios. By baking randomness into our testing, we can ensure that we’re not just testing for known requirements and corner cases but for broader purposes. This way, there’s less chance that an easy case may trip up our design.
- Adversarial Testing. We provide our systems with input that is deliberately malicious and/or harmful. This is another form of “breaking” our system; it improves our technology by exposing failure points, allowing us to identify potential safeguards and mitigate risks.
The five proof forms serve to prove that the technology is robust. If the system can overcome random variables, exhaustion, and contradiction to a reasonable degree, its robustness and adaptability will be validated, affirming its readiness for real-world challenges. Our ability to define the problem and our strategy to validate the desired behavior gives us the confidence that a solution exists.
Our Multi-Faceted Validation Strategy
Our validation approach embraces a multi-faceted strategy, driven by multiple aspects:
- Requirement Driven. Our validation efforts are guided by specific requirements that align with the intended functionality of our self-driving truck. We design for the known variables and the known unknown variables.
- Design Driven. We systematically validate our technology’s design to ensure alignment with Formal and Mathematical methods, enabled by MBSE, and validate that the system design is confirmed by the implemented system.
- Scenario Driven. Our technology is tested across a spectrum of real-world scenarios, ranging from routine to novel situations. We carefully define our system boundaries to minimize the unknown unsafe.
- Data Driven. Empirical evidence from real-world mileage, test runs, simulations, and controlled environments provides a factual basis for assessing our technology’s performance. This also allows us to expose new unknowns, validate assumptions that we’ve already made, and ensure that our requirements are as complete as possible.
Driving the Future of Freight: Validation
Torc Robotics’ validation strategy reflects a comprehensive approach to tackling the challenges of self-driving truck technology. By meticulously defining problems, embracing diverse proof techniques, and adhering to a multi-faceted validation strategy, we are propelling the industry towards true Level 4 readiness. Anchored in safety management and engineering rigor, Torc Robotics is not only shaping the trajectory of self-driving trucks but also setting a precedent for responsible and robust autonomous vehicle development.
