Mar 19, 2026
“XAI as the Solution to the Autonomous Driving AI ‘Black Box’”... ACEWORKS CEO Seung-beom Park Presents Future Verification Strategy
[Sanup Ilbo] ACEWORKS presented an integrated software and hardware verification strategy for ensuring the reliability of AI-based autonomous driving at Automotive Testing Expo Korea 2026, held at KINTEX in Ilsan on the 18th.
Speaking at a concurrent conference, ACEWORKS CEO Seung-beom Park highlighted a fundamental challenge facing the industry: despite rapid advances in autonomous driving technology, the decision-making process of AI still operates as a "black box" — its inner workings invisible even to developers.
To address this, ACEWORKS proposed the application of Causal Inference technology, which moves beyond conventional data training to examine why an AI reaches a particular judgment in a given situation. Park argued that true explainability requires breaking down an AI's actions into three discrete stages — perception, judgment, and control — and rigorously analyzing the rationale at each step.
He likened the approach to how an MRI pinpoints the exact site of a suspected brain condition, underscoring the value of verification at the micro level.
On the exhibition floor, ACEWORKS showcased its ADSmart actuator platform, a series of devices designed to translate autonomous driving software decisions into physical actions — and verify that those actions perform as intended.
The steering actuator mounts directly onto the steering column as a self-contained unit, requiring no modification to the existing Electronic Power Steering (EPS) system. A key safety feature is its Override function, which continuously monitors driver torque and instantly hands control back to the driver when needed.
The brake pedal actuator takes a mechanical approach, using a servo motor to physically press the brake pedal. Mounted in the driver's footrest area, it is designed to stay out of the driver's way during normal operation.
Also on display was BMS HILS, a verification solution for high-voltage battery management systems. The system simulates battery packs at voltages up to 1,000V, generating ultra-precise individual cell voltage outputs at 1mV resolution to validate BMS control logic. Built around a linear-type battery cell simulator, it produces minimal switching noise and scales up to 256 channels — enabling real-time simulation of the full battery pack found in a production electric vehicle.
"Before the industry can meaningfully leverage Explainable AI (XAI), domestic players need to build up an extensive base of real-world driving data," Park said. "ACEWORKS supports that effort by supplying physical actuators and high-fidelity simulation environments that streamline data acquisition and help companies meet SOTIF, the international functional safety standard."
Source: Sanup Ilbo, Reporter Ji-won Lim jnews@kidd.co.kr
URL: https://kidd.co.kr/news/245322