This project aims to implement a neural network to predict smooth, feasible trajectories that navigate around obstacles ("rocks"). The model should learn to balance imitating human-like paths with a physical constraint to avoid collisions.
- Feature Vector: trajectory_start_x, trajectory_start_y, trajectory_end_x, trajectory_end_y, rock_loc_x, rock_loc_y, rock_radius
- Model Prediction: trajectory (num of points x 2)
- Loss Functions:
- Reconstruction Loss: This loss will ensure the predicted trajectory follows similar route as the actual trajectory for a given start and end location
- Collision Loss: This loss adds a hard constraint to the predcited trajectory, ensuring it does not collide with the obstacle
Training dataset consists of trajectories which are valid and not necessarily straight lines. Testing dataset might consists of trajectories which are in collision, except the start and end location of the trajectories.