Trajectory Prediction is the task of forecasting the future path of moving agents based on their past positions — essentially predicting where a pedestrian, car, or robot will be in the next few seconds to enable safe planning.

What Is Trajectory Prediction?

• Input: Past coordinates $(x, y)$ for frames $t-N$ to $t$.
• Output: Future coordinates for frames $t+1$ to $t+M$.
• Difficulty: The future is multimodal (a person could turn left OR right). Models must often predict a distribution of possible futures.

Why It Matters

• Self-Driving Cars: "Will that pedestrian cross the street in front of me?"
• Social Navigation: Robots moving through crowds without bumping into people.
• Sports: Predicting where a player is running to pass the ball.

Methods

• Social Forces: Modeling interactions (people repel each other like magnets).
• Social LSTM / Social GAN: RNNs that share hidden states to model group dynamics.
• Transformer: Attention mechanisms to model long-range temporal dependencies.

Trajectory Prediction is AI foresight — allowing autonomous systems to act proactively rather than just reacting to the present moment.