Autonomous driving holds great promise for easing the driver shortage issue that has occurred in many countries. In this paper, we consider a semi-autonomous transit system where (1) individual buses could form platoons to increase bus capacity during peak hours and operate separately during off-peak hours; and (2) buses may drive autonomously in certain geofenced areas yet need to be guided by human drivers elsewhere (e.g., SAE level 4, semi-autonomous). We model buses driving autonomously within the geofenced area as short-turn bus service, and jointly optimize bus dispatch headway, platoon size and service type through an integer programming model to address the driver shortage issue. The proposed model considers real-world and micro-level bus operation processes for mixed traffic of short-turn and full route buses, with bus fleet and driver workforce constraints being modeled endogenously. Since the model is non-linear, we further develop a radial basis function-based surrogate framework to solve the model efficiently. Experimental results show that, compared to the conventional bus service that uses fixed bus capacity only, semi-autonomous bus platooning service reduces operation cost significantly while reducing passenger wait time. System performance variations of the proposed model under different driver availabilities are also examined.