Katkat
Back

Design and Debug an Automotive Sensor and Camera-Blockage System

RivianSoftware Engineer · Technical Screen
MediumSoftware EngineerTechnical ScreenSystem Design

Last updated

Tests your ability to design an automotive perception sensor layout and lens-blockage detection system, define safe vehicle degradation strategies, and systematically troubleshoot intermittent false-positive diagnostic alerts in the field.

Scenario

You are an embedded software engineer on an autonomous driving team designing the vision and diagnostic architecture for a next-generation electric utility vehicle. The vehicle supports Level 2+ highway pilot (lane centering, adaptive cruise control, automatic lane changes) and low-speed automated maneuvering (parking assist, tight trail navigation).

During operation, external camera lenses are subject to environmental hazards such as mud splatter, snow, dust, direct sun blinding, and condensation. You must design a robust sensor layout, an on-device blockage detection subsystem, and a graceful degradation manager. Additionally, you must troubleshoot an active field issue where vehicles intermittently log a false-positive "Front Wide Camera Fully Occluded" fault approximately once per hour during clear daytime driving.

Requirements

Functional Requirements

  • Sensor Selection & Placement: Propose camera and complementary sensor (e.g., radar, ultrasonic) locations to ensure full 360-degree coverage with redundancy for forward highway driving.
  • Blockage Detection Pipeline: Implement a real-time health-monitoring mechanism to classify lens status (Clear, Partially Occluded, Fully Occluded, Blinded/Glare).
  • Degraded Behavior & Safety Fallbacks: Define how the vehicle's driving assistance features degrade when a critical sensor is degraded or lost (e.g., fail-operational vs. fail-safe transitions, driver handoff timing).
  • Telemetry & Event Snapshotting: Capture lightweight diagnostic snapshots (sensor metadata, synchronized timestamps, low-resolution context frames) when health status changes.

Non-Functional & Embedded Constraints

  • Compute & Latency: Blockage detection must run on an automotive SoC edge accelerator consuming < 5% of perception compute budget, flagging full occlusion within 1.0 second.
  • Bandwidth: Telemetry uploads over cellular cannot exceed 20 MB per driving hour during normal operations.

Troubleshooting Task

A fleet technician reports that a batch of validation vehicles exhibits a repeatable diagnostic anomaly: the front-facing wide camera triggers a Fully Occluded diagnostic event roughly once every 60 minutes (+/- 2 minutes) under bright, clear daylight conditions, causing brief disengagements of highway steering assist.

Explain how you would collaborate with the technician to:

  1. Formulate testable root-cause hypotheses.
  2. Safely reproduce and isolate the issue without risking test drivers.
  3. Identify necessary telemetry, logs, and synchronized bus traces (e.g., CAN, Ethernet, PTP timestamps).
  4. Validate a software/firmware fix to eliminate regressions before broad deployment.

Hints

Hint 1

Consider combining classical computer vision heuristics (e.g., spatial frequency, edge gradient loss, histogram variance) with cross-sensor validation (e.g., comparing camera-detected optical flow against vehicle speed and radar targets).

Hint 2

Degraded vehicle states should enforce a deterministic fallback hierarchy: retain longitudinal control via radar if forward vision fails, while issuing an immediate take-over request for lateral control.

Hint 3

For the hourly false positive, consider periodic time-triggered embedded events: memory buffer wrap-arounds, PTP/NTP clock synchronization jumps, automatic exposure recalibration routines, or thermal management fan cycles.

Follow-up questions

  • How would you differentiate between uniform low-contrast environmental conditions (e.g., driving through dense fog or a dark tunnel) and physical lens blockage like mud?
  • How should the on-vehicle diagnostic engine manage high-rate ring buffers so that raw pre-incident camera frames are saved without saturating eMMC flash write cycles?
  • If an active lens-cleaning system (washer jet / air puff) is available, what logic and hysteresis thresholds would you use to trigger it without obstructing the driver's view?