From Automation to Physical AI: The Technical Destination Freight Has Been Building Toward

by Walter Grigg | Jun 18, 2026 | AI, Education, Featured

For decades, the trucking industry has chased the same goal—move more freight, more reliably, at lower cost, safely—driving successive waves of investment in route optimization, telematics, electronic logging, and mechanized loading. But every wave shared a fundamental characteristic: it was designed to eliminate variability, not navigate it.

That distinction matters more than it might first appear—and understanding it explains why physical AI is not simply the next technology upgrade for freight. It is the destination the industry has been moving toward all along.

The Ceiling That Was Always There

Traditional automation works by reducing the number of decisions a system needs to make. Define the conditions, define the response, let the system execute. In structured environments—a fixed production line, a mechanized fulfillment center—this is extraordinarily effective.

Long-haul freight is different. U.S. trucks move over 72% of all domestic freight by weight across a $900 billion annual system running through inherently unpredictable conditions: shifting weather, variable traffic, tight delivery windows, regional infrastructure differences, and a workforce navigating real demographic and economic pressures that have no quick resolution.

Automation can make a skilled driver more efficient. It cannot provide the judgment skilled driving requires. That is the ceiling… and freight has been pressing against it for years.

What Changes with Physical AI

Physical AI is not a new concept arriving from outside the industry. It is the progression that emerges when AI design moves from pattern-matching and rule-following to environment-understanding and adaptive decision-making.

Where traditional automation asks “what condition matches this rule?”, physical AI asks “what is happening here, and what should I do about it?” Systems built on physical AI perceive their environment in real time, interpret changing conditions, and make decisions based on live inputs, not pre-programmed responses.

For trucking, that distinction is the difference between a system that helps manage a route and a system that can drive one.

This is why physical AI isn’t simply an upgrade to existing automation. It is the technical foundation for an entirely different class of operations, one where the system carries the judgment, adaptability, and situational awareness that freight has always demanded. And critically, it extends the operational envelope well beyond what any rule-based system can reach, giving freight networks the resilience and adaptability their scale has always required.

Where Deployment Starts and Where It Goes

Understanding how physical AI enters freight requires one key concept: the operational design domain, or ODD, the defined set of conditions in which a system operates reliably, and from which it expands. This is not a limitation. It is how serious technology deployment works, and it maps naturally onto how the freight industry itself is structured.

Early physical AI applications are concentrating on the conditions where they perform most reliably: high-volume interstate corridors, predictable long-haul lanes, and structured handoff environments—also, not coincidentally, the segments under the greatest cost and capacity pressure. Efficiency gains in fuel consumption, vehicle utilization, and operational uptime compound quickly across large fleets.

From that foundation, the envelope expands. As systems accumulate operational experience and capabilities mature, physical AI moves into more variable conditions, broader geographies, and more complex logistics environments. The trajectory follows the same arc as every major technology adoption the freight industry has seen: prove it in the highest-value, most predictable application first, then scale.

The Fuller Picture

Autonomous trucks on long-haul corridors are the most visible expression of physical AI in freight, but only one part of a broader shift across the entire logistics chain. Adaptive freight handling systems, warehouse robotics, intelligent fulfillment automation, and predictive maintenance platforms all reflect the same underlying change: systems that understand objectives, not just instructions, increasingly capable of operating in the environments that have always resisted full automation: dynamic, variable, and high-stakes.

What this means for freight over the next decade is a transition from fragmented, single-purpose tools to integrated, intelligent platforms. Fewer systems doing more. Greater resilience across supply chains. And a coherent answer to the operational complexity that has defined freight’s constraints for decades.

The Destination, Not a Detour

It would be easy to frame physical AI as another technology wave arriving to disrupt an industry that moves deliberately. That framing misses what’s actually happening.

The trucking industry has been building toward this for decades. Every automation investment—every route optimizer, every telematics platform, every mechanized dock—was a step toward more efficient, more reliable, more adaptive operations. Physical AI is not a detour from that path. It is where the path was always leading.

The industry that moves eleven billion tons of freight each year has always needed systems capable of operating intelligently in a dynamic, variable, real world. Those systems now exist. The question is no longer whether physical AI belongs in freight. It’s how quickly the industry builds on the foundation it has already spent decades laying.

Digital AI vs. Physical AI: What’s the Difference and Why It Matters

by Kiley Thompson | Apr 15, 2026 | Education, Featured, Technology

Have you used AI to streamline your work or generate a list for you? Have you been fooled by an AI image? The technology we’ve been watching in movies for 20 years is now in our homes, cars, phones, and schools. Artificial intelligence is no longer theoretical. Most people—across industries and age groups—interact with tools like ChatGPT, Claude, or Gemini. Work apps like Zoom now offer AI summaries, and Microsoft and Adobe have generative AI tools built into their products. This widespread exposure has created a shared baseline: AI is useful, real, and improving quickly.

But there’s a critical distinction that’s still not widely understood—digital AI vs. physical AI.

Digital AI:

Intelligence On-screen

Digital AI operates in a purely virtual environment. It processes text, images, and data to:
• Answer questions
• Generate content
• Analyze patterns
• Support decision-making

Its rapid rise wasn’t accidental. Digital AI benefited from a massive, ready-made dataset: human language and images. For decades, we’ve been writing, storing, and digitizing information—emails, documents, books, photos, websites. It’s been trained on the Internet: the largest repository of language and visuals in human history.

Once that data became accessible, AI systems could be trained quickly and at scale.

That’s why digital AI feels like it “appeared overnight.” It was trained “in the cloud” of ever-expanding data centers, allowing for rapid iteration.

Physical AI:

Intelligence in the Real World

Physical AI takes the same foundational technologies as digital, on-screen AI—machine learning and neural networks—and applies them to real-world interactions.

Instead of predicting the next word, physical AI must decide:
• What object am I seeing?
• How do I move around safely?
• How hard should I grip this item?
• What action should I take next?

This introduces massive complexity. Instead of just structured data, physical AI must work within the real, unstructured environment, around people and things that don’t always follow the rules. The real world isn’t just data—it’s dynamic, unpredictable, and only governed by physics.

Physical AI needs to account for the 3D space of dimensions, with sensors, actuators, and an understanding of interactions and their outcomes.

Why Physical AI Is Slower to Scale (for Now)

When we built digital AI systems, we chose the right dataset by only using some parts of the internet. Chatbots were built using mostly social media data; Coding Agents were built using mostly open source code data.

Physical AI also requires special data for real world use cases. The world is extremely complex, and the sensors that “see” it can also be complex. But, this time, we can’t just look on the Internet to “find” it. The sensors we use, the environments in which robots move, and the special situations we have to prepare for have simply not been recorded.

Unlike digital AI, physical AI lacks a rich historical dataset like the Internet. There’s no ready made data set for how humans move through environments, how objects behave under force, or how tasks are physically completed. All the physical data must be created, not just collected.

The same generative tools that you use in digital AI can be applied within a simulated world, using real-world, grounded physics, and the physical AI hardware (robots and sensors) can be built alongside the AI. We don’t always have to wait for a robot to go out into the world, record the data, and bring it back to us – the robot isn’t ready yet! Companies like NVIDIA, Tesla, and Amazon are accelerating this by combining compute power, simulation, and real-world data collection.

At Torc, our autonomous driving system, TorcDrive, is being trained to work in the real world using both real-world data (recorded camera and lidar images of on-road driving) as well as complex simulated images created from those same on-road recordings, simultaneously.

The Key Insight: Same Brain, Different Body

The most important takeaway is this: digital AI and physical AI are built on the same core technology. Digital AI is proving what’s possible and adaptable by humans, and most importantly, helpful. Physical AI extends that capability into the real world. Everything you’ve seen AI do on a screen—learning, adapting, improving—will eventually happen in the physical world.

What This Means for the Freight Industry

For freight, logistics, and operations, digital AI helps companies and people think better … physical AI can help you perform better.

At Torc, physical AI is manifested in our autonomous driving software, TorcDrive, and powered by AV 3.0, on our trucks today, after being trained by millions of hours of real world and simulated scenarios. It’s the realization of the promises of hundreds of years of technology and human invention. We are still in early days of building physical AI but the trajectory is clear. Digital AI and physical AI aren’t separate revolutions. They are one and the same.

Meet Your First Physical AI

A physical AI machine you might be familiar with is the in-house robot vacuum. The simplest vacuum has the most basic sensors to register the world around it. If it bumps into something, the computer registers the bump against the moveable panel and turns the wheels to rotate itself in a different direction. Other more expensive models can use sensors to determine floor material, and then adjust how they clean accordingly.

More Going On Under the Hood

There are robot vacuum models with self-emptying functions, pet sensors, and even self-cleaning controls. One of the newest models introduced at CES 2025 even has a robotic arm to pick up socks. As we transition to a more physical AI world, how physical AI “understands” and interacts with its surroundings is very different “under the hood” (or dust bin, in this analogy). It’s becoming more complex, and smarter, and importantly, more applicable and helpful.

Not All Physical AI Thinks the Same

How different physical AI instances and machines “think” and how they must act on their sensors vary widely. It is critical to keep in mind that systems are running ever-increasing AI models, trained on real world and simulation data, designed to interact with the physical world in more capable ways, far exceeding the scalability (and capability) of the first physical AI technologies.

Freight Industry Physical AI Key Applications and Workflows

Autonomous Warehouse Operations
Intelligent Fleet Management & Safety
Dynamic Load Optimization
Automated Material Handling
And of course … Autonomous Vehicles

AV 3.0: Torc’s AI Blueprint

by Torc | Aug 6, 2025 | Education

The autonomous ready Freightliner Cascadia, with Torc virtual driver software embedded within the chassis.

Torc has set our sights on one goal: to create a scalable product capable of delivering the immense value of autonomy to our freight partners. The system, hardware, and software must all be built for safety, adaptability, and cost-efficiency; and work seamlessly on Daimler Truck’s Freightliner Cascadia.

And Torc has cracked the code: we’ve designed an unparalleled system that unlocks safe and reliable autonomous driving for long-haul trucking, which can be quickly scaled to meet the needs of our fleet customers. We call it AV 3.0.

AV 3.0

To hit the high goals we set, we recognized that today’s most advanced AI approaches being used to develop autonomy today would not meet the level of capability, performance, and most importantly, safety that the trucking industry requires. We knew we needed to achieve several firsts for autonomous vehicles in parallel: a new autonomous driving architecture as well as a new data and development infrastructure.

Torc tapped our world-class AI talent and our long history in robotics and self-driving vehicles to create AV 3.0: our virtual driver software, the advanced data loop, and generative AI simulation infrastructure to train and test it, setting ourselves apart from others in the industry.

Let’s break down the parts and explain the advanced components of each.

A graphic describing the three parts of Torc's AV 3.0

The Software Product: Virtual Driver

Virtual driver, our autonomous software product driving the truck, is comprised of three main functional modules: perception, prediction, and planning. It is an end-to-end reinforcement-learned (RL) modular software stack with redundant rule-based (heuristic) guardrails. This allows us to uniquely deliver the optimal mix of autonomous driving performance, system reliability, and driving safety. While the machine learning modules in the stack provide the highest driving performance output, Torc layers in rule-based models and arbitration as redundant guardrails, ensuring the ML model doesn’t violate the basic rules of driving.

The addition of the heuristic guardrails is one of the benefits of Torc’s 20-year heritage in developing autonomous vehicle software products. Building rule-based models is an intensive, manual, and time-consuming process, and Torc has been working with heuristic rules as part of our code since the beginning. Also, the addition of rule-based models is inherently verifiable and interpretable since the rules are hand-coded – another layer of verifiability to support a strong safety case.

The modular architecture of the stack provides fine-grained introspection: the ability to see and validate with transparency and explainability if all the learned and heuristic modules are reacting properly to the input. In comparison, an AV 2.0 end-to-end system is a “black box” with just sensor data in and trajectory out. You can’t understand how the stack is reasoning, only that it is reasoning. There’s no way to check on the intermediary results to check if the module is behaving properly and therefore, no way to effectively debug issues. Every time an issue is found, an AV 2.0 end-to-end system must be retrained as a whole with new data specific to that issue to achieve a correct output, but there is no way of knowing if it truly fixed the issue. This “whack-a-mole” approach adds significant time and uncertainty to the software creation and safety validation process.

With our AV 3.0 stack, we can understand how the system is reasoning within each module, in perception, prediction and planning, and where the problems are, retrain the stack with the right set of generated data to specifically target the issue, and supervise the module-level output to validate the fix.

Perceived object bounding boxes and lane centers from the perception module.

The Simulation Environment: Learning to Create the World

Supporting the training of our virtual driver is our world generation and data-driven simulation environment, comprised of several important and complex pieces.

It Always Starts with the Data. We capture real-world data with the actual production-intent sensor suites on our trucks and have collected a breadth of highway and surface street data across many regions and driving conditions.

In fact, Torc has by far the largest set of real-world driving data in autonomous trucking, spanning many diverse scenarios and corner cases.

We then recreate that real-world data within our data-driven simulation environment, giving us the ability to replay those scenes ad infinitum to test the virtual driver. But that real-world data, used repeatedly on replay, is also the starting point for the generative AI environment we’ve developed, to create infinite sequences of challenging driving scenes and scenarios necessary to train and test the virtual driver to meet the needs of our ODD and safety goals.

As no such capability to generate the realistic data needed was available in the market, the Torc AI team leveraged years of research and development to deliver these capabilities, providing us with a unique time, cost, and data fidelity advantage.

In fact, Torc has by far the largest set of real-world driving data in autonomous trucking, spanning many diverse scenarios and corner cases.

These can be summarized as the following three capabilities:

Neural Rendering

A generative AI approach for realistic scene creation. Based on real world scenes, we can recreate freight-specific, customer-desired sections of road with various “actors” — cars, trucks, animals, pedestrians — and then generate unique additions and/or changes to the scene, like a car braking hard in front, a wrong way driver, a light snow shower that becomes a blizzard, in a photo-and-lidar-realistic output that can be injected into our perception stack.

World Simulation

This approach generates the other reactive objects that react to our truck (the ego vehicle) and then manages the behaviors however we want or need. We can make reactions more or less aggressive on all of the actors, creating infinite scenes and behaviors. For example, one car captured behaving normally on one section of road can then be made to behave aggressively by degrees. It can go from benign to aggressively merging into the truck’s lane in various places by millimeters and timings by milliseconds.

Importantly, all learned behaviors are grounded in real-world physics-based models, so while we could test with Hollywood-style simulations of flying cars or people, we stay grounded in actualities. We instead simulate vehicles skidding into our lane where in the real-world data, they had been driving carefully in their own lanes.

scenario generation

Scenario generation takes world simulation to the next level, generating never-before-seen road geometries, creating all road conditions, all edge cases, and repeating the simulations through millions of miles, allowing for an even more diverse set of road scenarios. This allows us to quickly adapt the virtual driver software to any new route with minimal additional data collection and time, dramatically accelerating our ability to scale our software to drive new highways, new autonomous hub layouts, in different states, and so on.

Neural Rendering cropping

Real world footage

“For fully data driven simulation, you need to be able to simulate the behaviors and do so very realistically. You also need to be able to control the reactions, in a data-driven way,” says Felix Heide, Head of AI at Torc, and one of the architects of Scenario Dreamer™, a recent diffusion method for scenario generation. “This combination of world simulation systems accomplishes what is necessary.”

The Proof: Training and Testing

Using these layers of world simulation, neural scene rendering together with generated sequences with fully reactive agents, we repeatedly fine-tune and deeply test the virtual driver, as mentioned previously, an instructor teaching a “behind-the-wheel” student, by using reinforcement learning, to explore decision making and provide feedback at the same time.

Virtual driver is taught not only by providing the ideal road trajectory, but also by providing additional scene information so that the system learns to explicitly predict and, on top of that, to explicitly use the information for its final trajectory decision. This, in return, allows a more accurate training process, which, as a byproduct, helps us understand in more detail how the entire system behaves. That detail can then be fed back into the system.

We apply all these techniques to create the data needed to deeply test the virtual driver behavior for the most unusual or unlikely corner cases, to give us high confidence of safe behavior on the road, i.e., that the training of our AI system and the virtual driver has been successful. After that confidence threshold is reached, we move the updated model out of the simulated environment and onto the trucks, validating on both closed course and public roads for every software release.

It’s this high-velocity AV 3.0 framework of data –> training –-> testing –> release that sets an industry benchmark in software development and deployment.

The End Result: Safe and Scalable

“End-to-end systems are critical for scalability. It allows us to be much more adaptable to new needs or to fix issues quickly, and because we have a modular transparent stack, we can see the issues independently,” says C.J. King, Torc Chief Technology Officer. “We can use our generative AI stack to generate the data to train the outcomes rather than train a camera-based object and distance detector, or lidar and radar object detection separately from the prediction and planning. It’s another layer of getting the technology ready for the road.”

This high velocity is critical not only to keep to our release schedule, but to ensure we can quickly enable our customers to scale to new routes and hubs. We can identify, triage, and deploy issues to engineering before we ever encounter them on the road. Additionally, issue resolution and fixes are done within days, not weeks. This turnaround means there can be multiple fixes and releases daily.

Why We Call It AV 3.0

Why do we call it AV 3.0? Torc’s innovation is the next step beyond the technology in AV 1.0 and AV 2.0, as we knew those approaches alone weren’t enough to succeed in creating and deploying autonomous trucks in a timely manner. Torc’s AV 3.0 drives the state of the art in the evolution of scalable, practical, and applied autonomous vehicle software.

Until we reach our product launch in 2027, we will continue to hone this technology, gather more real-world data (something we will always need), develop commercialization plans and hub designs, close our robust safety case, and solve a multitude of other necessary autonomous trucking puzzles. But AV 3.0 provides a clear path to how it can be done.

So, what’s next?

We’ve defined AV 3.0, the groundbreaking technology used to create, train, and test the virtual driver. What our virtual driver runs on, and how, is another industry-defining element setting Torc apart from the competition. We’re deploying this technology right now, not on prototypes or demonstration trucks, but on the trucks we will go to market with… on production-intent embedded hardware.

Next Up: “Scaling Physical AI for Autonomous Trucking Today: Torc’s First-to-Production Embedded Hardware”

Transparency

AV 3.0 understands the state of the entire model and why it acted the way it did, both within individual modules and how it affected the entire end-to-end system. There’s visibility and understanding of what’s always going on inside the model at all times.

Explainability

AV 3.0 permits traceability and verifiability, providing us with the ability to see why and where our virtual driver did something. We can understand both what is detected and what reasoning the system made to determine the object and its meaning. This is crucial for the explainability of our end-to-end system and for our safety case.

Improved Accuracy

Introspection means the AI model is aware of its internal state, how it makes decisions, and its behaviors. Models can understand why they are doing something and can therefore be more accurate.

Let’s explain EXPLAINABILITY and IMPROVED ACCURACY together: AV 3.0 provides object detection outputs and information on the 3D world view our vehicle is in – Is that a car? At what velocity are other highway users travelling at? Is that an emergency vehicle? Where is the lane center? A traffic light? This level of detail allows us to understand what the virtual driver thinks in a certain situation, and we also know how it acts with that information.

Improved Safety

With full transparency of the entire model and individual modules, it is easier to monitor model behavior and verify the software from the inside and explain perception and decision making in full.

Torc’s First Responder Program Recent Highlights

by Torc | Jun 26, 2025 | Education

Torc’s First Responder Team recently conducted comprehensive training sessions for more than 150 first responders representing 33 distinct agencies along the Texas I-35 corridor. This initiative is a critical part of Torc’s commitment to preparing the first responder community for interacting with Torc’s self-driving trucks on public roads.

Richard Russell, Senior Manager of First Responder Policy, and Foster Murphy, Fleet Compliance Specialist — with a combined 40+ years of law enforcement experience across city, county, and state levels – delivered training on Torc’s trucks to first responders from various Texas agencies, including Fort Worth PD, Waco PD, Williamson County Sheriff’s Office, and Johnson County Sheriff’s Office.

The sessions provided a deep dive into the essentials of autonomy and offered a hands-on experience with the Torc autonomous truck. Over two hours, Richard and Foster guided first responders through the operational mechanics of autonomous commercial motor vehicles, emergency response protocols, and safety procedures to ensure first responders are well-equipped to handle various scenarios involving Torc trucks.

Looking Ahead

Torc plans to expand its First Responder Training to additional agencies, focusing on where Torc trucks will be operating along I-35 in Texas. Torc is dedicated to fostering trust and collaboration with first responders, recognizing their critical role in ensuring public safety as autonomous technology becomes more prevalent. By continuing to partner with state and local agencies, Torc is committed to advancing the safe adoption of self-driving commercial vehicles, paving the way for a future where innovative technology and public safety go hand in hand.

Check out our First Responder page for more information.

Securing the Future: Meet Michael Maass, Director of Product Cybersecurity

by Alex Mesa | Jun 5, 2025 | Cybersecurity, Education

At Torc, safeguarding safety-critical systems from evolving cybersecurity threats isn’t just a priority—it’s a core
mission.

At the helm of this mission is Michael Maass, the Director of Product Cybersecurity and Principal Product
Cybersecurity Architect, whose career reflects a deep dedication to building secure technologies, strong teams,
and forward-thinking strategy.

A Career Rooted in Cybersecurity Excellence

Michael brings over 20 years of cybersecurity experience to the table—17 of those spent specifically on
securing companies, products, and safety-critical systems. His background bridges the technical and strategic:
from writing low-level software in x86 assembly, C/C++, and Java, to leading cross-functional security teams
and developing ways to build security into products.

Throughout his career, Michael has helped companies adopt cutting-edge practices that meet both technical and
compliance requirements. His passion lies in one of the most complex and important areas of modern
technology: developing secure, safety-critical cyber-physical systems that can stand up to real-world threats
and regulatory scrutiny.

Building Secure Teams and Culture

Michael’s leadership extends beyond technology. He’s spent years building and nurturing top-tier cybersecurity
teams in the automotive space, ensuring they not only have deep technical skills but also understand the
broader impact of their work. His ability to communicate with executives, regulators, and external stakeholders
makes him a rare bridge between engineering and compliance—a necessity in today’s high-stakes
cybersecurity environment.

Leading Cybersecurity at Torc Robotics

At Torc, Michael wears two hats: he acts as both Principal Product Cybersecurity Architect and the Director of
Product Cybersecurity. In these roles, he leads efforts to embed cybersecurity into every stage of product
development for autonomous vehicles—systems that must operate safely and securely in the unpredictable real
world.

His leadership ensures that cybersecurity is an integral part of a product’s full lifecycle, ranging from inception
to product retirement.

Penetration Testing: Red Teaming for the Right Reasons

One of the key tools in Michael’s cybersecurity arsenal is penetration testing—a technique that simulates real-
world attacks in order to find and fix vulnerabilities before bad actors can find and exploit them.

“Penetration testing is essentially where someone with hacking skills applies those skills to a particular target,
with the goal of efficiently finding vulnerabilities and identifying hardening opportunities,” Michael explains.

Penetration testing is a process within the A Versatile Cybersecurity Development Lifecycle (AVCDL), an open
source document set crafted by Charles Wilson, Cybersecurity Architect at Torc, Michael, and cybersecurity
engineers at Torc and other companies, for use by the autonomous vehicle industry and any other creating
safety-critical cyber physical systems. This structured lifecycle ensures that every aspect of the product, from
hardware to software, is rigorously vetted for cybersecurity risks and those risks are addressed.

Penetration testing is just one part of a broader cybersecurity strategy. While penetration testing is often
spotlighted because it’s exciting and accessible, it’s important to note that a secure cybersecurity platform
includes a comprehensive set of practices. Many of these, while equally critical, are less known unless you’re
deeply immersed in the domain.

Still, penetration testing stands out as an illustrative example of how Torc’s structured lifecycle ensures that
every aspect of an autonomous vehicle product—from hardware to software—is rigorously vetted for
cybersecurity risks.

Inside the Penetration Testing Process

As Michael says, no two penetration tests are the same, but most follow six general steps:

Pre-Engagement: Define the target, set objectives, and establish boundaries.
Reconnaissance: Gather data on the system using both passive and active techniques.
Threat Simulation: Emulate the tactics of real-world threat actors.
Exploitation: Attempt to breach the system, identifying weak points.
Analysis: Assess findings and potential business impact.
Reporting & Recommendations: Share results and collaborate on mitigation strategies.

In a real-world example of a penetration test on a lidar system, Michael shared that safety and security go
hand-in-hand. Both components must work together to ensure the strength of a system. During this particular
lidar test, Michael pointed out that analog attacks, while harmful, aren’t necessarily as impactful as system-level
exploits.

Michael’s Vision for Secure Autonomy

Michael’s ultimate goal is to create technology that’s secure, reliable, and compliant—all while enabling
innovation. His work ensures that Torc’s systems are ready not just for today’s challenges, but that the larger
autonomous vehicle industry is ready to face tomorrow.

As Michael says, a rising tide lifts all boats. At Torc, that philosophy is core to how cybersecurity is
approached—not just as a competitive advantage, but as a shared responsibility across the industry. That’s why
Torc is pushing forward with tools like the AVCDL. By making this framework visible and accessible, Torc aims to
help everyone build safer, more secure, and more compliant products.

“I’m passionate about developing secure, safety-critical systems that balance innovation with acceptable
liability and compliance,” he says. In an industry where trust is everything, that mission is more vital than ever.
With decades of experience and a passion for securing the future of mobility, Michael Maass is helping to
shape the next generation of cybersecurity in autonomous vehicles. Through strategic leadership, technical
expertise, and a commitment to continual improvement, he’s ensuring that Torc stays ahead of the curve—
keeping systems safe, secure, and ready for the road ahead.

Advancing Safe Machine Learning: “The Community Now Owns This”

by Torc | May 29, 2025 | Education

At the recent SAE World Congress, Torc took the stage to share something big: a new safety approach to using machine learning (ML) in high-stakes areas like self-driving trucks. Paul Schmitt, Torc’s Senior Manager for Autonomy Systems, presented a paper called “The ML FMEA: A Safe Machine Learning Framework.” The work, co-authored with experts from Torc and safety partner TÜV Rheinland, addresses a major challenge in using AI for safety-critical applications: how do you know the AI is safe?

Machine learning models are often described as “black boxes”—it’s hard to see how they make decisions, and that makes it hard to ensure they’re making the right ones. As Schmitt explained during the talk, existing safety standards highlight the importance of managing risk but don’t give clear, practical tools for how to do it. That’s what inspired the team to create the ML FMEA.

ML FMEA stands for Machine Learning Failure Mode and Effects Analysis. It builds on a well-known tool, FMEA, that industries have used for decades to catch potential problems before they happen. Torc and its partners adapted this trusted method to fit the unique challenges of machine learning systems—like those used in autonomous trucks.

What makes this approach special is how it brings two very different groups—machine learning engineers and safety experts—into the same conversation. “My favorite benefit is that it gives both teams a shared language to understand and reduce risk,” Schmitt said. The framework helps teams walk through each step of the ML process and think through what could go wrong, why it might go wrong, and how to prevent it.

The team didn’t stop at the idea—they created a working template to help others put the approach into action. It includes real examples of possible failures and how to fix them, from the moment data is collected to the time the ML model is deployed and monitored in the real world.
And in the spirit of industry collaboration, Torc and TÜV Rheinland made the framework public. “We see this as a first step toward safety-certified machine learning systems,” Schmitt said. “These challenges don’t just affect self-driving trucks. They affect healthcare, manufacturing, aerospace—you name it. So we open sourced the method and template, and we’re excited to see how others improve it.”

Partnership

Schmitt also highlighted the importance of partnership: “We were thrilled to work with TÜV Rheinland on this project. Bodo Seifert instantly brought depth and credibility to the work.”

The presentation drew strong interest, with attendees snapping photos of slides and downloading the paper on the spot. During the Q&A, co-authors Krzysztof Pennar and Bodo Seifert joined Schmitt on stage to take questions. “We heard great ideas on how to expand the approach from automakers, safety experts, and standards committee members,” said Schmitt. “Seeing that level of engagement—especially from the standards community—was honestly a dream come true.”

The paper was co-authored by Bodo Seifert, Senior Automotive Functional Safety Engineer at TÜV Rheinland, Jerry Lopez, Senior Director of Safety Assurance; Krzysztof Pennar, Principal Safety Engineer; Mario Bijelic, AI Researcher; and Felix Heide, Chief Scientist.

As AI becomes more common in critical systems, tools like ML FMEA will be key to making sure it’s not just powerful—but also safe.

Register Today

for The ML FMEA Presentation Virtual Event

On Wednesday, June 11, 2025, at 1pm ET / 10am PT, four of the authors of The ML FMEA: A Safe Machine Learning Approach will be presenting on this paper as well as fielding questions. Find out more and register for the Virtual Event by visiting the presentation page.

register for the virtual event