https://speakerdeck.com/furuhashifumihito 2026-04-12 11:48:59 -0400 Sun, 12 Apr 2026 00:00:00 -0400 https://speakerdeck.com/furuhashifumihito/ibaraki-seminar-for-resilient-and-future-city-number-1 https://speakerdeck.com/furuhashifumihito/ibaraki-seminar-for-resilient-and-future-city-number-1 We study interaction-aware trajectory planning for autonomous vehicles in dense traffic. Standard MPC treats surrounding trajectories as fixed and can become conservative or inconsistent. We model planning as a noncooperative differential game and approximate the Nash solution with a learned surrogate for online use. The surrogate encodes ego and neighbors (MLP+GNN), predicts dynamics with an LSTM, and outputs controls refined by a safety layer that enforces collision-avoidance and road-boundary constraints. Physical consistency is ensured by embedding the dynamic bicycle model directly into the optimization, preserving vehicle dynamic feasibility. Gradients through the safety layer are propagated via a straight-through estimator. To address nonconvex, multi-objective training, we adopt evolutionary model merging (DARE-TIES). In synthetic multivehicle scenarios (1–10 cars, 5-s horizon), our method lowers instantaneous cost versus MPC and iterative DG baselines and yields stable, collision-free trajectories. The framework offers a practical, real-time path to physically consistent, interaction-aware planning. We study interaction-aware trajectory planning for autonomous vehicles in dense traffic. Standard MPC treats surrounding trajectories as fixed and can become conservative or inconsistent. We model planning as a noncooperative differential game and approximate the Nash solution with a learned surrogate for online use. The surrogate encodes ego and neighbors (MLP+GNN), predicts dynamics with an LSTM, and outputs controls refined by a safety layer that enforces collision-avoidance and road-boundary constraints. Physical consistency is ensured by embedding the dynamic bicycle model directly into the optimization, preserving vehicle dynamic feasibility. Gradients through the safety layer are propagated via a straight-through estimator. To address nonconvex, multi-objective training, we adopt evolutionary model merging (DARE-TIES). In synthetic multivehicle scenarios (1–10 cars, 5-s horizon), our method lowers instantaneous cost versus MPC and iterative DG baselines and yields stable, collision-free trajectories. The framework offers a practical, real-time path to physically consistent, interaction-aware planning. Sun, 12 Apr 2026 00:00:00 -0400 https://speakerdeck.com/furuhashifumihito/hksts-outstanding-student-paper-award-physically-consistent-differential-game-surrogates-for-interaction-aware-av-trajectory-planning https://speakerdeck.com/furuhashifumihito/hksts-outstanding-student-paper-award-physically-consistent-differential-game-surrogates-for-interaction-aware-av-trajectory-planning This paper introduces the Differential-Game based Physics-Informed Neural Network (DG-PINN), a trajectory-prediction model that embeds a multi-agent differential-game formulation into a physics-informed neural network. By treating each vehicle as a strategic agent, DG-PINN simultaneously captures inter-vehicle interactions and individual utilities, thereby unifying longitudinal car-following and lateral lane-changing behaviors within a single, interpretable framework. Physical laws derived from the Hamilton-Jacobi-Bellman equations are incorporated as loss regularizers, enabling the model to retain the data-efficiency and interpretability of physics-based methods while leveraging the representational power of deep learning. We evaluate DG-PINN on the NGSIM I-80 highway dataset under varying training-sample regimes (20-100 sequences) and compare it with constant-velocity/acceleration baselines, a Physics Uninformed Neural Network (PUNN), and an IDM-based PINN. DG-PINN achieves the lowest root-mean-square error, average displacement error, and final displacement error across all sample sizes, outperforming IDM-PINN and PUNN by up to 19 % in low-data settings while exhibiting markedly lower variance. Moreover, the learned utility parameters allow quantitative profiling of driver traits—such as safety-mindedness and aggressiveness—and visualize the rationality of observed versus hypothetical maneuvers. The results demonstrate that integrating differential-game theory with PINNs not only improves prediction accuracy and robustness but also yields actionable insights for explainable decision-making in autonomous-vehicle systems. This paper introduces the Differential-Game based Physics-Informed Neural Network (DG-PINN), a trajectory-prediction model that embeds a multi-agent differential-game formulation into a physics-informed neural network. By treating each vehicle as a strategic agent, DG-PINN simultaneously captures inter-vehicle interactions and individual utilities, thereby unifying longitudinal car-following and lateral lane-changing behaviors within a single, interpretable framework. Physical laws derived from the Hamilton-Jacobi-Bellman equations are incorporated as loss regularizers, enabling the model to retain the data-efficiency and interpretability of physics-based methods while leveraging the representational power of deep learning. We evaluate DG-PINN on the NGSIM I-80 highway dataset under varying training-sample regimes (20-100 sequences) and compare it with constant-velocity/acceleration baselines, a Physics Uninformed Neural Network (PUNN), and an IDM-based PINN. DG-PINN achieves the lowest root-mean-square error, average displacement error, and final displacement error across all sample sizes, outperforming IDM-PINN and PUNN by up to 19 % in low-data settings while exhibiting markedly lower variance. Moreover, the learned utility parameters allow quantitative profiling of driver traits—such as safety-mindedness and aggressiveness—and visualize the rationality of observed versus hypothetical maneuvers. The results demonstrate that integrating differential-game theory with PINNs not only improves prediction accuracy and robustness but also yields actionable insights for explainable decision-making in autonomous-vehicle systems. Sun, 12 Apr 2026 00:00:00 -0400 https://speakerdeck.com/furuhashifumihito/dg-pinn-differential-game-based-physics-informed-neural-network-for-vehicle-trajectory-prediction https://speakerdeck.com/furuhashifumihito/dg-pinn-differential-game-based-physics-informed-neural-network-for-vehicle-trajectory-prediction