Autonomous Systems

Navigation

Autonomous drones use computer vision, SLAM, and AI for navigation. These systems can be attacked through sensor spoofing and adversarial inputs.

Autonomous Navigation Systems

Visual Navigation

  • • Optical flow (position hold)
  • • SLAM (mapping)
  • • Obstacle avoidance
  • • Landing detection

Attack Vectors

  • • Camera blinding
  • • Pattern spoofing
  • • Adversarial patches
  • • Sensor flooding

Vision System Attacks

vision-attacks.sh
bash
# Attacks on visual navigation:

# 1. IR blinding (DJI sensors)
# IR LEDs can saturate downward sensors
# Causes altitude hold failure

# 2. Optical flow confusion
# Moving patterns under drone
# Causes position drift

# 3. Obstacle detection spoofing
# Fake obstacles to block path
# Or mask real obstacles

# 4. Landing marker spoofing
# Fake AprilTags or landing markers
# Redirect precision landing

# 5. SLAM manipulation
# Modify visual features
# Confuse loop closure detection

# ROS-based drone simulation
roslaunch px4 mavros_posix_sitl.launch
# Test attacks in simulation first