Katkat
Back

Design Human Avoidance for Warehouse Robots

AmazonSoftware Engineer · Technical Screen
MediumSoftware EngineerTechnical ScreenSystem Design

Last updated

This problem evaluates your ability to design a real-time sensor-to-actuator pipeline for an autonomous robot, focusing on latency, sensor fusion, and safety-critical path adjustments.

Problem: Autonomous Robot Obstacle Avoidance

You are tasked with designing the 'Safety & Avoidance' subsystem for a warehouse robot that operates in a dynamic environment with moving human workers and other robots. The robot is equipped with a 360-degree LiDAR, ultrasonic proximity sensors, and wheel encoders.

Requirements:

  1. Functional Requirements: The system must detect humans in the path, calculate a safe stopping distance, and trigger an emergency stop (E-stop) or trajectory adjustment if a collision risk is detected.
  2. Latency: The round trip from sensor input to motor controller output must be under 50ms.
  3. Reliability: The system must account for sensor noise and potential occlusion of human subjects.

Your Design Should Cover:

  • Sensor Fusion Strategy: How do you reconcile data from LiDAR (high resolution, long range) and ultrasonic sensors (short range, low resolution) to confirm an object is a human?
  • Data Flow: Describe the pipeline from raw sensor data acquisition to the motor command interface.
  • Safety Protocol: Define the 'Decision Logic' for when to slow down versus when to execute an emergency stop based on robot velocity and obstacle proximity.
  • Failure Modes: How does the system behave if the primary LiDAR sensor reports a 'dirty lens' or stops sending data?

Assumptions:

  • Robot max speed: 2 m/s.
  • The robot has a pre-loaded map of the warehouse.
  • You have access to a middleware layer (e.g., ROS2-like) that handles message passing between nodes.

Hints

Hint 1

Consider implementing a 'Safety Envelope' around the robot that expands dynamically based on the current velocity vector.

Hint 2

Use a Kalman Filter or similar estimation technique to track moving obstacles and predict their trajectory over the next 500ms.

Hint 3

Prioritize a 'Fail-Safe' state where the system defaults to a controlled stop if the heartbeats of critical sensor processes are not detected.

Follow-up questions

  • How would you optimize the perception pipeline if the robot encounters a 'crowded' area with 20+ detected objects simultaneously?
  • How would you integrate a machine learning model for human pose estimation without violating the 50ms latency constraint?
  • If the robot detects an obstacle but the floor is wet, how should the braking logic change?