The field of autonomous driving is rapidly advancing, with a focus on improving the accuracy and efficiency of trajectory prediction and planning. Researchers are exploring new approaches to model complex interactions among agents, such as formulating interactions as a level-k game framework and utilizing metrics like Trajectory Entropy to quantify game status. Another area of research is the development of lightweight and end-to-end trainable architectures, such as Coupled Convolutional LSTM (CCLSTM), to capture temporal dynamics and spatial occupancy-flow correlations. Additionally, flow matching-based motion prediction frameworks like TrajFlow are being introduced to address scalability and efficiency challenges. Diffusion models are also being improved, with variants like Fast Monte Carlo Tree Diffusion offering significant speedups. Furthermore, researchers are working on enhancing collision avoidance capabilities with the environment, using contrastive learning-based modules like ECAM. Notable papers include: Trajectory Entropy, which refines the level-k game framework through a simple gating mechanism, and CCLSTM, which achieves state-of-the-art performance on occupancy flow metrics. TrajFlow, which predicts multiple plausible future trajectories in a single pass, and Fast Monte Carlo Tree Diffusion, which offers a 100x speedup over standard MCTD. ECAM, which significantly reduces collision rates when integrated with existing trajectory forecasting models.