§01 · Current Focus
Two Forces in Mixed Traffic
Behavioral Consensus & Multi-Objective AV Learning
Autonomous vehicles must balance safety, efficiency, and interaction, goals that are frequently at odds and cannot be resolved by any single hand-crafted reward. This thread defines behavioral consensus as the joint alignment of these three dimensions, derives an empirical Pareto surface directly from naturalistic trajectories, and uses that surface as the target a reinforcement learning agent, trained with Soft Actor-Critic in SUMO, learns to drive toward on a timestep basis, with no access to future data.
Quantum-Inspired Representation of Driver Heterogeneity
Rather than reducing driver heterogeneity to discrete labels such as aggressive or cautious, this thread represents each surrounding driver as an evolving density matrix in a nonlinear feature space, continuous, probabilistic, context-dependent, and history-dependent. The representation recovers interpretable driver profiles without supervision and reproduces macroscopic traffic phenomena, such as the fundamental diagram and hysteresis, that were never built into the model.
