Mohammed Amine Bencheikh Lehocine

Mohammed Amine Bencheikh Lehocine is a PhD candidate at Mercedes-Benz AG in collaboration with the Munich University of Applied Sciences. His research is on end-to-end models that perceive and predict at the same time: instead of connecting a detector and a trajectory predictor through hand-crafted bounding boxes — which throws information away and lets errors propagate — his models detect the road users around the vehicle and forecast where they will go in one fully differentiable network.

His first contribution, MASAR, follows the idea of “looking backward to look forward”: for every detected object the model reconstructs its past trajectory, gathers the appearance features along that path and lets the appearance evidence decide which motion hypothesis is right. The long-term temporal context obtained this way improves the forecast of future trajectories by more than 20 % on nuScenes while keeping the detection quality. The work was accepted at ICRA 2026. He studied at Université Paris Dauphine and is passionate about AI and data-driven solutions for autonomous driving.