Why attackers care about this surface
Drones and robots collapse several security domains into one operational stack: firmware integrity, flight-control logic, telemetry exposure, remote management, mission planning, companion-computer trust and the human operator in the loop. A weakness in one layer can turn into loss of navigation integrity, unsafe movement, sensor spoofing, denied mission execution, operator deception or direct control-path compromise.
In practice that means the assessment has to move across layers. You may start at a radio link or a management interface, pivot into a companion computer, then validate what that means for the autopilot, the vehicle state and the mission outcome. The key question is always the same: what commands, assumptions or safety gates can an attacker influence from the outside?
Primary cyber-physical attack paths
Telemetry & control links
Command, telemetry and mission messages whose authentication, replay resistance and routing determine whether control remains trustworthy.
Companion computers
Onboard systems that bridge sensors, networks and flight control, often expanding a service foothold into vehicle-level access.
Autopilot & firmware trust
Firmware provenance, update paths, configuration integrity and command validation at the flight-control layer.
ROS 2 / robot middleware
ROS 2 nodes, topics, services and DDS security settings that shape discovery, authorization and message trust.
Sensors & navigation inputs
GNSS, vision, inertial and range inputs whose spoofing or corruption can turn valid control logic into unsafe movement.
Mission tooling & operator workflow
Ground-control software, planning files, credentials and update practices that connect human decisions to vehicle commands.
How to approach the assessment
Start by drawing the control chain end to end: ground station, radio path, base station, relay services, companion compute, autopilot, sensors and actuators. Then identify where commands originate, how trust is established, which channels are authenticated, what is signed, what is merely accepted and which faults degrade into unsafe but still “valid” behavior.
On mature targets, spend time on protocol visibility and message boundaries rather than only on web or host findings. On smaller fielded platforms, the fastest wins often sit in configuration inheritance, exposed serial interfaces, weak maintenance habits or over-trusted integration scripts.
What success can look like for an attacker
Control-path manipulation
Unauthorized or altered commands that change vehicle state, mission behavior or actuator decisions.
Telemetry deception
False position, state or health data that misleads automation and human operators.
Companion pivot
Movement from an exposed onboard service into trusted middleware, telemetry channels or flight-control interfaces.
ROS graph exposure
Unprotected discovery, topics or services that expose robot state or permit unauthorized message publication.
Operator workstation compromise
Loss of the planning or ground-control endpoint that holds mission data, credentials and trusted vehicle access.
Safety bypass by integration drift
Individually valid components whose combined configuration weakens interlocks, fails closed-loop assumptions or permits unsafe commands.
Reference links
- PX4 DocumentationAutopilot architecture, companion computer integration, simulation and developer workflows.
- MAVLink Developer GuideProtocol reference for telemetry, commands, missions and system interaction.
- ArduPilot Developer DocumentationOpen autopilot stack for aerial, ground and marine vehicles.
- ArduPilot ROS 2 IntegrationROS 2 coupling and development notes for ArduPilot-based systems.
- PX4 ROS 2 User GuideROS 2 architecture and middleware integration with PX4.
- ROS DocumentationRobotics middleware, nodes, topics and ecosystem-level references.
Where to pivot when the vehicle is only part of the story
Network
Internal-service discovery, segmentation, credential paths and lateral movement across supporting vehicle infrastructure.
Advanced
Specialist research spanning cloud, identity, mobile, APIs, AI, OT/ICS and other non-routine assessment surfaces.
AI Security
Model and agent security for perception, planning and autonomy features that can influence robotic decisions.