trajectory prediction

**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.