As camera-based DMS reaches maturity, attention is shifting to what it still can’t see. In this interview, we explore with Vered Levy-Ron, CEO of CorrActions, the limitations of gaze-based systems, the challenge of validating cognitive-state detection, and why integrating driver readiness into L2+ and L3 systems requires a fundamental rethink of how engineers define safety.
An Interview With:
Vered Levy-Ron CEO
Camera-based DMS has matured around gaze and head pose. From an engineering perspective, what critical safety gaps remain without cognitive-state insight?
From an engineering perspective, the primary safety gap in current monitoring systems is the latency between cognitive decline or change in cognitive state and physical manifestation; camera systems are fundamentally reactive, triggering alerts only after visible symptoms like yawning or drooping eyelids appear, by which point reaction times are already dangerously compromised. These systems suffer from a blind spots regarding cognitive distraction and intoxication, where a driver may maintain a correct gaze and head pose while experiencing disconnection from the driving experience. Without the ability to detect early-stage neurocognitive states such as slowed cognitive processing, the current technology holds blind spots in its ability to predict and prevent accidents.
Your solution leverages existing in-vehicle signals rather than dedicated hardware. How do you ensure signal reliability and robustness across different vehicle platforms and architectures?
To ensure reliability across diverse vehicle architectures, we employ a two-pronged approach centered on deep collaboration and continuous system diagnostics.
Strategic OEM and Tier-1 Collaboration: Because every vehicle architecture is unique, we work hand-in-hand with OEMs and Tier-1 suppliers right from the initial phase. We clearly define our data requirements upfront and collaborate to map their existing network data to our solution, ensuring the vehicle supports and delivers exactly what we need.
Continuous Runtime Monitoring: Once deployed, our software will actively monitor the health and integrity of incoming signals in real-time. If our system detects an anomaly such as a signal dropout or corrupted data it immediately logs and reports the issue. This constant monitoring ensures that any degradation in data quality is handled safely and effectively, providing a hardware-free solution that is both highly reliable and scalable.
What does a validation framework look like for cognitive-state detection, particularly when aligning with safety requirements and emerging regulations?
CorrActions is proud to announce the successful completion of our latest validation phase, spanning both closed-track and open-road environments. Highlighted by a rigorous blind test conducted by a major OEM, our technology demonstrated near-zero false positives while achieving exceptional recall.
Validating a software-only, neuroscience-based system like our NeuroMonitor requires bridging neuroscience and physical AI with automotive safety. To meet stringent safety standards and emerging regulations like EU GSR2 and upcoming Euro NCAP protocols, our framework relies on four core pillars:
- Ground truth based -CorrActions correlate our kinematic data (micro-muscle movements on the steering wheel) with medical-grade benchmarks, such as EEG readings, KSS and BAC. This ensures our system is scientifically anchored to actual brain activity.
- Controlled impairment testing (simulators & closed tracks): Because testing severe impairments on public roads is dangerous, we use a two-step approach. We safely test in advanced driving simulators, and then validate those findings on closed tracks. This allows us to train our AI on real-world vehicle physics and steering vibrations without risking public safety.
- On-Road testing: We validate our algorithms against massive datasets of everyday, unscripted driving across diverse vehicles and road conditions. This ensures normal behaviours (like reaching for a drink or hitting a pothole) aren’t falsely flagged, guaranteeing high accuracy and driver trust.
- Regulatory & functional safety alignment: As an AI-driven software solution, we align our development with rigorous automotive functional safety standards to ensure our system behaves predictably and safely, even in complex, real-world edge cases.
As vehicles transition toward L2+ and L3, how should engineers think about integrating cognitive readiness into takeover strategies and system-level decision-making?
Integrating cognitive state detection requires engineers to treat the driver as a dynamic component of the vehicle’s safety rather than a binary switch. In L2+ and L3 systems, the primary danger is the cognitive disconnect: a state where the driver remains physically positioned for control but is mentally absent due to automation complacency.
To bridge this gap, engineers must implement adaptive takeover logic that scales based on the driver’s cognitive state. If CorrActions NeuroMonitor detects low cognitive distraction, the system cannot assume a standard 10-second handover is safe. Instead, the vehicle’s decision-making engine must proactively widen safety margins such as increasing following distances or reducing speed to compensate for the driver’s delayed reaction time.
At InCabin USA, what specific technical challenges or questions are you most interested in discussing?
At InCabin USA, the conversation must evolve beyond simple gaze-tracking to address the industry’s most pressing technical blind spot – The driver’s cognitive state. While camera-based DMS has reached a level of maturity in identifying physical distraction, the engineering community is still grappling with the cognitive gap.
The Validation of Invisible ground truth: As we move toward detecting impairments like cognitive distraction and early-onset fatigue, traditional benchmarks fall short. The next challenge in the automotive industry is how to bridge the gap between clinical, medical-grade physiological data and the unpredictable, noisy environment of dynamic on-road behavior. Establishing a standardized ground truth for cognitive states is the next step for safety certification.
Architecting the Multi-Modal Safety Net: Gaze vectors and head poses are only proxies for attention; they don’t capture the “mind-wandering” that occurs during semi-autonomous driving. I’m looking to discuss the integration of multi-modal sensor fusion, specifically how software-driven physiological insights-such as kinematic micro-movement tracking-can complement traditional optical data. The goal is to move from reactive systems to a robust, persistent safety net that understands the driver’s mental “Time to Reach” even when their eyes are technically on the road.
By shifting the focus from where a driver is looking to how they are processing information, we can finally mitigate the false sense of security inherent in Level 2+ systems and ensure a safe, synchronized handover of control.
Don’t miss Vered’s full presentation ‘Decoding the Driver’s Brain: Differentiating Cognitive Impairments through Physical AI for High-Confidence Intervention‘ on the InCabin Exhibition Stage, Thursday 11th June at 1:55pm EDT.
Interested in exterior vehicle sensing technology?
With a pass to InCabin USA, you’ll also get full access to our
co-located sister event, AutoSens. Find out more here >>
