How do autonomous systems predict future motion of surrounding objects?

Motion prediction estimates where other road users will be in the future. Approaches include: Physics-based models (constant velocity, constant acceleration, bicycle model for vehicles); maneuver-based prediction (recognizing intentions like lane change, turn); interaction-aware prediction (considering how objects influence each other); and deep learning methods (RNNs, transformers trained on trajectory data). Prediction horizon typically extends 3-8 seconds. Output often includes multiple hypotheses with probabilities (multimodal prediction). Challenges include: Pedestrian behavior randomness, intent recognition, and handling rare maneuvers. Prediction quality directly impacts planning safety, with conservative predictions improving safety but potentially reducing efficiency.