The technology behind Silvereye.
Silvereye is developing one core software system, Pegasus. This page explains what it is intended to do, how it fits alongside a drone’s existing systems, and where development stands.
PEGASUS
Autonomous intelligence software for unmanned aerial systems.
In developmentAutonomous intelligence for the next generation of flight.
Pegasus is Silvereye’s developing software system, designed to improve how compatible unmanned aerial platforms perceive their surroundings, estimate their position and navigate challenging operating environments.
Its initial engineering focus is GPS-resilient positioning and navigation. Further capabilities are planned as the system develops.
Mission-level intelligence
- Estimates position and velocity from camera and inertial data
- Understands the surroundings and plans routes
- Reports confidence and system health to the operator
- Sends estimates and setpoint requests through MAVLink or ROS 2
Safety-critical flight control
- Stabilises the aircraft and drives the motors
- Validates every request it receives from Pegasus
- Keeps its own failsafes, independent of Pegasus
- Remains the final authority over the aircraft
Pegasus is designed to run alongside a compatible platform’s existing flight controller, not to replace it. No general-purpose language model controls the aircraft, and nothing in Pegasus drives the motors directly.
How the components fit together.
Pegasus is organised into four capability areas that sit between the operator and the platform. Safety and Reliability runs alongside the others rather than at the end of the chain.
Sets intent and limits. Supervises, and can intervene at any time.
Environmental Perception
Turning camera data into an understanding of the surroundings the aircraft can act on.
Perception extracts what matters from the camera stream: stable visual features for navigation, nearby obstacles, and features relevant to the mission, such as a structure being inspected. It feeds both the navigation estimate and higher-level decisions.
- Computer vision
- Environmental understanding
- Obstacle awareness
- Recognition of relevant environmental features
Supplies features to Navigation, and obstacle and scene information to Mission Intelligence.
Mission Intelligence
Interpreting a mission and choosing how to fly it as conditions change.
Mission Intelligence turns an operator’s intent into routes and decisions. It works at mission level, above the flight controller. Instead of replaying a fixed list of waypoints, it should take account of navigation confidence and what Perception reports, then continue, adjust or hand back to the operator.
- Mission interpretation
- Route planning
- Navigation decisions
- Adaptation to changing environmental conditions
Takes intent from the operator, state from Navigation and scene information from Perception. Sends route and setpoint requests to the flight controller, within limits set by Safety and Reliability. It does not control motors directly.
Safety and Reliability
Knowing how much to trust the system at any moment, and keeping a human in charge.
Safety and Reliability runs alongside every other layer. It watches confidence values and system health, enforces operating boundaries, and makes sure the operator can see what the system is doing and take over. Confidence reporting is designed into the navigation work from the start.
- Confidence monitoring
- Fault detection
- System health monitoring
- Human supervision and intervention
- Safe operating boundaries
Monitors all layers. Can restrict Mission Intelligence, alert the operator, or request a predefined safe action from the flight controller.
An illustrative sequence.
One example of how the layers are intended to respond when conditions change. It illustrates the design, not a tested result.
- 01Condition changes
The aircraft flies beneath a bridge deck and satellite signal quality falls.
- 02Safety and Reliability
Confidence monitoring sees that GNSS no longer agrees with the visual-inertial estimate.
- 03Navigation Intelligence
GNSS is weighted down. Position continues from camera and inertial data, with growing uncertainty reported.
- 04Mission Intelligence
Given that confidence, the mission continues, slows, holds or returns, within limits the operator set.
- 05Human operator
The operator sees the change in source and confidence, and can take control at any point.
Built on open robotics infrastructure.
Rather than starting from nothing, Silvereye intends to build on established open-source tools, so it can work with existing platforms and the engineering teams that use them.
PX4
Open-source flight control. Silvereye is designed to work alongside it rather than replace it.
ROS 2
Robotics middleware for connecting perception, estimation and planning components.
MAVLink
The standard protocol for exchanging state and commands with the flight controller.
Visual SLAM
Building a map of the surroundings while locating the aircraft within it.
Visual-inertial odometry
Estimating motion by fusing camera images with inertial measurements.
Simulation and validation
Developing and testing in simulated environments, then validating against recorded data, before any hardware is flown.
What the architecture is built around.
Confidence first
Every estimate comes with a measure of how far it can be trusted. Decisions use that measure, and operators can see it.
Platform-compatible
Software for existing, compatible platforms through open interfaces, not a new airframe or proprietary autopilot.
Human oversight
Autonomy within boundaries a person sets, with clear ways to supervise and take over.
Help develop Pegasus.
If you build platforms, run operations where positioning is unreliable, or work on visual-inertial navigation, we would like to compare notes.