Characterizing lane-changing behavior in mixed traffic
This study analyzed 7,636 lane-changing events from the Waymo Open Motion Dataset using a game-theoretic framework to model interactions between active lane-changing vehicles and passive target-lane vehicles. Using k-means clustering and a quantal response equilibrium framework, the researchers classified vehicle behaviors as cooperative or defective and constructed empirical payoff tables. The analysis revealed that social dilemmas occur in 4% of active and 11% of passive lane-changing scenarios, primarily manifesting as Stag Hunt or Prisoner's Dilemma games.
Why it matters — It demonstrates that automated vehicles exhibit higher rates of cooperative behavior than human-driven vehicles in both active and passive lane-changing roles, and proves through simulations that repeated traffic interactions drive long-term cooperation regardless of automated vehicle market penetration.
Caveat: The findings are based on trajectory data from a single curated autonomous vehicle dataset, which may not fully represent the diverse driving cultures and conditions of general mixed traffic.