Self-driving features from companies such as Tesla are marketed as a path to hands-free travel, but a series of recent accidents has highlighted a key reality: today’s autonomous systems can still misread fast-moving, high-risk situations.
A group of scientists now suggests that the next major safety improvement might come from a surprising place - the minds of the people sitting inside the vehicle.
In research published in Cyborg and Bionic Systems, a team in China explored whether tracking passengers’ brain activity could guide autonomous cars toward safer choices when danger arises.
The researchers relied on a noninvasive method called functional near-infrared spectroscopy (fNIRS). This technique monitors brain signals associated with stress, emotional responses, and perceived risk in real time.
Lead author Xiaofei Zhang of Tsinghua University explained that fNIRS can provide insight into how humans experience risk and emotion, making it a promising way to support smarter autonomous driving. The team developed a decision-making model that incorporates passengers’ physiological states to boost both safety and efficiency when vehicles encounter hazardous scenarios.
Their system integrates this brain-based information directly into an autonomous vehicle’s control software. When higher levels of stress or perceived danger are detected in riders, the vehicle automatically adopts a more cautious driving style.
The approach uses deep reinforcement learning, allowing the car to adapt more quickly and select safer actions by learning from human reactions as well as environmental data. During testing, the system shifted into a conservative mode whenever passengers appeared uncomfortable, helping the vehicle respond more carefully in potentially dangerous conditions.
Results showed improvements over conventional autonomous driving strategies, including faster learning, enhanced safety, and greater passenger comfort.
The authors emphasized that the work is still preliminary. The driving situations simulated were relatively basic, and participants represented only a small, similar demographic group, which limits how broadly the findings can be applied.
Future studies, the researchers said, will test the method in more complex, real-world environments and combine brain-based information with more vehicle sensor data to improve risk detection even further.
Newer Articles
- Long-Awaited Gmail Update: Users Can Swap Out Old Addresses
- Deepfakes of the Departed: The New Battleground Over Likeness and Legacy
- Robotaxis, Safety, and the ‘Pink Tax’: Waymo’s New Year’s Eve Challenge
