Despite recent advances in autonomous driving, most self-driving systems primarily emphasize short-term trajectory prediction for motion planning, often neglecting the critical role of future 3D vehicle orientation (posture) in achieving safe and stable navigation—especially in dynamic and cluttered environments. Accurate posture forecasting is vital for anticipating vehicle behavior, enabling smoother control, and avoiding unsafe maneuvers. In this work, we present PosFormer, a novel transformer-based architecture that predicts future vehicle orientation by jointly leveraging visual perception and motion intent. PosFormer extracts rich spatial features from six in-vehicle camera views using a convolutional backbone, and aligns them with planned trajectory information through a cross-attention mechanism within a Transformer decoder. This fusion enables precise, ego-centric 3D orientation forecasting in quaternion form at future horizons (e.g., 3 and 6 steps ahead). Unlike traditional approaches that decouple perception from planning or rely on heuristic filters, PosFormer offers a unified, data-driven framework that captures context-aware motion dynamics. Experiments on real-world autonomous driving datasets demonstrate that PosFormer achieves average quaternion angular errors of \(0.87^\circ \) , \(0.79^\circ \) , and \(2.26^\circ \) across future frames.