[Feature]: Incooperate for Non-Linear Modelling for Trajectory Prediction

[simoooong]

Feature Description

Currently, trajectories for both target objects and the ego-vehicle are modeled using simple linear extrapolation of velocity. This approach fails to account for non-linear dynamics such as acceleration, braking, and angular velocity (turns).

The TrackingInfo module currently only stores historical data and basic motion data without considering yaw rates. This leads to inaccurate collision predictions, especially in curves or during speed changes.

Note that TrackingInfo does not calculate any of these values, and just stores the historic data, as well as motion data (which again does not account for any angular velocity) and has to be incooperated.

Furthermore carla provides an OpenDrive map which may be integreted for trajectory prediction (e.g. detected vehicle on right turn lane, its trajectory is most likely being a right turn)

Definition of Done

• TrackingInfo updated to calculate acceleration/braking/angular velocity
• Trajectory prediction logic accounts for non-linear effects
• Ego-vehicle future pathing accounts for current acceleration/braking state in the collision check.
• Collision check logic updated to use a more robust intersection calculation than the current "closest point" heuristic.
• Consider for map-based trajectory gating using OpenDrive information.

Testability

Unit Tests: Validate non-linear state transitions (e.g., verify the predicted path follows a curve when angular velocity is present).

Visual Testing: Validate Trajectories at intersections/curves

Dependencies

No response