To clarify the visual cognitive characteristics of drivers when passing through the access points of urban underground roads, the Jiefangbei Underground Ring Road in Yuzhong District, Chongqing was selected as the test site. A Tobii Glasses 2 wearable eye tracker was used to collect eye movement data from 30 drivers when passing through straight-segment, curved, and combined access points. Six key eye movement indicators, including saccade speed, saccade amplitude, pupil area, pupil diameter, pupil area change rate, and blink frequency, were screened using the CRITIC weighting method. An entropy-weighted TOPSIS model was then constructed to comprehensively evaluate the visual load of drivers at different types of access points. The results show that: (1) curved access points impose the greatest visual load on drivers, with an average pupil diameter of 4.618 mm and an average pupil area of 16.748 mm2, both higher than those of straight-segment and combined access points; (2) straight-segment connecting road access points exhibit the optimal visual characteristics, achieving the highest comprehensive evaluation score of 0.666 by the entropy-weighted TOPSIS method, with an average saccade speed of 220.962 px·s?1 and an average saccade amplitude of 54.256 px, both outperforming other types; (3) curved left-side access points cause the most significant visual load, with the lowest comprehensive evaluation score of 0.436, and their pupil area shows the largest individual differences; (4) combined access points have the highest pupil area change rate (1.718%), reflecting the greatest visual load fluctuation. These results reveal the influence mechanism of the types and layouts of underground road access points on drivers' visual load, and can provide theoretical support for optimizing access point design and improving driving safety.