This study investigates travel mode choice with on-demand autonomous vehicle (AV). It takes Singapore as the study area and specifically focuses on understanding the impacts of built environment (BE) on first-mile scenarios. A dynamic stated preference survey is developed to automatically generate first-mile travel scenarios based on the respondent’s dwelling location and real-world traffic information. Two mixed logit models with panel data structures are adopted to explore the impacts of BE on AV mode choice. The results reveal that BE factor is independent of trip specific and sociodemographic variables. Although including BE does not significantly improve the model fitting, it adds to explain the nuances of individual’s preference on travel mode choice. We then employ the models to forecast AV mode choice of 11,545 individuals from the Household Interview Travel Survey of Singapore. It provides a good understanding of AV market share in different areas and helps in evaluating the planning for AV deployment in Singapore. The estimation shows the mode shares of AV in most of pilot areas with AV plans are quite low. Thus, revisiting the planning of pilot areas for future AV deployment may be needed to avoid a spatial mismatch of on- demand AV service.