Pegasus Simulator

Photorealistic multirotor simulation on NVIDIA Isaac Sim with PX4-in-the-loop

License

BSD-3-Clause

Overview

Pegasus Simulator is a framework built on top of NVIDIA Isaac Sim / Omniverse for simulating multirotor vehicles with photorealistic rendering and accurate physics. Developed at ISR-Lisbon (Instituto Superior Técnico), it provides PX4 software-in-the-loop integration out of the box — a simulated drone flies with the real autopilot stack and QGroundControl — plus a Python API for custom controllers, sensors and RL-style experiments.

Pegasus wraps Isaac Sim's PhysX physics and RTX rendering in an aerial-robotics-shaped API: vehicles are defined with rotor models (thrust curves, drag, rolling moments), sensor suites (IMU, GPS, barometer, magnetometer, camera) and communication backends. The flagship backend is PX4 SITL — Pegasus launches and connects the actual PX4 autopilot per vehicle, so control firmware, EKF2 estimation and QGroundControl mission flows behave exactly as on hardware, closing the classic sim-to-real gap at the autopilot level. A pure-Python control backend supports custom controllers and learning experiments without PX4, and ROS 2 interfaces expose the simulation graph. Multi-vehicle scenarios, USD-based photorealistic environments, and people/moving-obstacle utilities round out the research feature set. Published as an open framework with an accompanying paper, it has become the standard choice for labs that want Isaac-class rendering for aerial autonomy work where AirSim's discontinuation left a gap.

Why it matters

Pegasus occupies the spot AirSim left behind for serious aerial research: photorealistic, GPU-accelerated simulation with the REAL PX4 autopilot in the loop — the combination drone labs need for perception-heavy autonomy and sim-to-real work on modern hardware.

Problem it solves

Aerial autonomy research needs photorealistic sensors AND flight-accurate vehicle behaviour. Game-engine sims lack autopilot fidelity; flight-dynamics sims lack rendering. Pegasus runs PX4 itself against Isaac Sim physics and RTX rendering, giving both at once.

Key takeaways

• Built on NVIDIA Isaac Sim: PhysX physics plus RTX photorealistic rendering
• PX4 software-in-the-loop per vehicle — real autopilot, EKF2 and QGroundControl flows
• Python API for custom rotors, sensors, controllers and multi-vehicle scenarios
• ROS 2 interfaces and USD-based environments for perception research
• From ISR-Lisbon with an accompanying open publication

Who it is for

• Aerial robotics researchers
• PX4 developers testing autonomy in simulation
• Perception engineers needing photorealistic drone imagery
• RL practitioners for aerial control

Limitations

• Requires an RTX-class NVIDIA GPU and the heavyweight Isaac Sim installation
• Multirotor-focused — no fixed-wing or VTOL vehicle models in the core
• Tracks Isaac Sim releases, so version pinning matters