Safety-aware traffic assignment model in mixed CAV-HDV networks: a multi-player perspective

Fully controllable connected and automated vehicles (CAVs) have been extensively studied as potential tools for improving traffic efficiency, even coexisting with selfish human routing behaviors. However, their potential advantages when balancing efficiency and safety in route choice remain underexplored. Moreover, these advantages depend critically on operational objectives: selfish CAVs (privately owned or transportation network company (TNC)-operated pursuing individual or fleet cost minimization) may exacerbate system costs, while government-operated CAVs pursuing system optimum may improve system performance. To address these issues, this study develops a safety-aware mixed traffic assignment problem (SM-TAP) model. The model first extends the classic travel time cost by incorporating a route-based crash risk cost (CRC) considering both links and intersections, forming a generalized travel cost with non-monotonic properties in mixed traffic flow. The SM-TAP model is then constructed within a general variational inequality (VI) framework that models HDVs under stochastic user equilibrium (SUE) principles and CAVs according to their different operational objectives—individual travelers as user equilibrium (UE) players, TNCs as fleet optimization (FO) players, and traffic planners as system optimal (SO) players. The SM-TAP model and proposed path-based solution algorithm are evaluated through numerical experiments on the Nguyen-Dupuis, Sioux Falls networks and Changsha network, and benchmarked against a mixed traffic assignment model that considers only travel time. The results yield three main findings: (i) Regardless of CAV objectives, low CAV penetration rates (approximately 10%–20%) increase system costs when safety is considered, after which costs decline as CAV penetration increases; by contrast, when only travel time is considered, costs decrease monotonically with CAV penetration. (ii) When CAVs operate under the SO objective, system costs are consistently lower than under other perspectives, as routing choices can reduce HDV costs and ultimately improve overall efficiency, even at the expense of CAV costs. Conversely, “selfish fleets” (i.e., FO) behaviors impose a greater burden on the system than individually selfish routing (i.e., UE). (iii) Although the addition of CAVs on a given road segment may increase travel time due to congestion effects, rational system-level deployment of CAVs can consistently reduce safety costs, thereby lowering overall transportation system costs.