L4: Prediction and Behavior Forecasting

Bridge: From Perception to Action

Learning Objectives

• Predict future trajectories of detected objects

• Design path planning algorithms for autonomous navigation

• Generate smooth, feasible trajectories

• Understand the complete planning pipeline

Outline

Part 1: Trajectory Prediction

Covers methods to forecast future trajectories of surrounding agents using both physics-based and learning-based approaches. Discusses uncertainty modeling, intent inference, and multimodal behavior prediction.

Part 2: Hierarchical Planning Architecture

Explains the three-layer architecture of mission, behavior, and motion planning. Establishes the relationship between high-level decisions and low-level trajectory generation.

Part 3: Path Planning Algorithms

Introduces classical and modern planning algorithms including graph-based methods (Dijkstra, A*), sampling-based techniques (RRT, RRT*), and optimization-based approaches. Discusses environment representation, collision checking, and efficiency trade-offs.

Part 4: Trajectory Generation and Selection

Focuses on generating dynamically feasible and smooth trajectories. Covers polynomial fitting, spline-based motion generation, and cost-function-based selection for comfort and safety.

Part 5: Integration and Real-World Challenges

Explores how trajectory prediction and planning integrate with perception and control systems. Reviews real-world deployment challenges such as computational latency, map inaccuracies, and dynamic obstacles.