New AI framework dCAP enables autonomous trucks to perceive their environment despite constantly shifting camera positions from trailer articulation

dCAP uses transformer networks with cross-view and temporal attention to continuously estimate the 6-degree-of-freedom relative pose between tractor and trailer cameras

The framework replaces static sensor calibration with dynamic pose prediction, improving 3D object detection when integrated with BEVFormer

Addresses real-world challenges in autonomous trucking caused by fifth-wheel joint movement, trailer flex, and occlusions that confound existing perception methods

Researchers introduced STT4AT, a CARLA-based benchmark dataset to evaluate autonomous truck perception systems