Design navigation-safety simulation parameters and experiments

Q: Design navigation-safety simulation parameters and experiments

This is a Machine Learning interview question from Zoox for Data Scientist roles. View the full question and solution on PracHub.

Q: How do I approach Machine Learning interview questions?

Machine Learning questions require understanding of core concepts and practice. PracHub provides solutions with explanations to help you master machine learning interviews.

Question

You must design a simulation-based evaluation for navigation safety of an autonomous agent. Be specific: a) Enumerate scene parameters to vary and justify each for safety relevance (e.g., map topology/lane geometry, traffic density/composition, other-agent behavior models and aggressiveness, initial conditions and spawn rules, pedestrian/cyclist flows, occlusions, weather/lighting, road friction/grade, sensor noise/latency/dropouts, perception errors, rare hazards like cut-ins/jaywalks/doorings). b) Define distributions or generators for these parameters; state how you will cover the space (domain randomization vs curated scenario library vs adversarial generation) and how you will ensure repeatability. c) Specify safety metrics and how to compute them: collision rate per mile, near-miss counts (e.g., TTC < τ), intervention rate, constraint violations (RSS/DRAC), comfort, and multi-objective aggregation. d) Propose an experiment plan that estimates very rare-event risk without bias: outline importance-sampling or CEM-based scenario weighting and write the unbiased risk estimator with weights; discuss variance control. e) Given a baseline collision probability ~1e-4 per mile, target 95% confidence with ±20% relative error: propose a sample-size calculation and any batching strategy under a 10 compute-hour budget. f) Describe how you would calibrate and validate sim-to-real (parameter fitting from logs, coverage tests, sensitivity/ablation, goodness-of-fit), and how you would detect OOD scenarios at evaluation time.

Design navigation-safety simulation parameters and experiments

Comments (0)