How predict vehicles’ turn direction at intersection?
Company: Waymo
Role: Data Scientist
Category: Machine Learning
Difficulty: easy
Interview Round: Onsite
At an intersection, there are **N vehicles stopped or moving slowly**. For each vehicle you have historical time-series data up to the current time:
- Position \((x_t, y_t)\)
- Velocity \((v_{x,t}, v_{y,t})\) or speed + heading
- (Optionally) lane/map context for the intersection
### Task
Build a model to predict each vehicle’s **intent** at the intersection:
- **Left turn**
- **Go straight**
- **Right turn**
Explain:
1. How you would define labels and handle ambiguity (e.g., last-second lane change).
2. What features / model class you would use (from simple baselines to sequence models).
3. How you would evaluate performance (metrics, slicing, calibration) and avoid leakage.
4. Key edge cases and failure modes.
Quick Answer: This question evaluates a data scientist's competency in time-series intent prediction at intersections, including label definition under ambiguity, trajectory and map-based feature representation, model class selection, and evaluation considerations such as leakage prevention.