Design an autonomous-vehicle simulator for testing how a driving model responds to road situations.
Clarify how the model connects to the simulator, what information the model receives, and what outputs the simulator should produce. Describe scenario setup, the simulation loop, model integration, and how you would evaluate the resulting behavior. State the fidelity and timing assumptions behind your design.
### What a Strong Answer Covers
- An explicit boundary between the model under test, simulated observations, control actions, and resulting vehicle behavior.
- Scenario construction, environmental variation, and reproducible runs.
- Synchronization between simulated time, sensor observations, and model execution.
- Evaluation of driving outcomes and the limits of conclusions drawn from simulation.
### Follow-up Questions
- How would you distinguish a model regression from a simulator or scenario change?
- What happens if model inference is too slow to meet the control deadline?
Overview: Design an autonomous-driving simulator with model interfaces, synchronized scenarios, reproducible evaluation, timing checks, and fidelity limits.
Design an autonomous-vehicle simulator for testing how a driving model responds to road situations.
Clarify how the model connects to the simulator, what information the model receives, and what outputs the simulator should produce. Describe scenario setup, the simulation loop, model integration, and how you would evaluate the resulting behavior. State the fidelity and timing assumptions behind your design.
What a Strong Answer Covers Guidance
An explicit boundary between the model under test, simulated observations, control actions, and resulting vehicle behavior.
Scenario construction, environmental variation, and reproducible runs.
Synchronization between simulated time, sensor observations, and model execution.
Evaluation of driving outcomes and the limits of conclusions drawn from simulation.
Follow-up Questions Guidance
How would you distinguish a model regression from a simulator or scenario change?
What happens if model inference is too slow to meet the control deadline?