Simulation Framework
DreamLander trains in an environment based on the six-degrees-of-freedom (6-DOF) equations of motion between the lander and its desired landing point. The environment is built in IsaacLab and uses its integrated NVIDIA PhysX engine for dynamics propagation. PhysX supports collision handling and full 6-DOF constraints, making it suitable for the simulation’s multiple sensors. The simulation focuses on the near-landing phase of entry, descent, and landing (EDL), so gravity is assumed to point directly down.
Terrain in IsaacLab is generated from FBX files designed and textured to resemble the colors and geometric features—such as craters, mountains, and rocks—of the planet or asteroid being simulated. High-fidelity terrain renders are important for photorealistic camera observations and can help reduce the simulation-to-reality gap. This work does not use pixel-based observations, so ultra-high-fidelity texture rendering is not required; accurate geometric terrain representation is.
The lander’s 3D model is another important part of the simulation framework. IsaacLab computes the vehicle’s inertia tensor from user-defined mass or density parameters. Choosing a lander with geometric features similar to a real vehicle improves simulation fidelity. The lander model has no texture because texture does not affect the simulation.
Frequently Asked Questions
What simulation framework does DreamLander use? It uses IsaacLab and its integrated NVIDIA PhysX engine to propagate dynamics.
What does the simulation model? It models 6-DOF motion between the lander and its desired landing point, focusing on the near-landing phase of EDL.
How is terrain represented? Terrain is generated from FBX files designed to resemble planetary or asteroid features, including craters, mountains, and rocks.
Why does the lander model’s geometry matter? IsaacLab computes the vehicle’s inertia tensor from mass or density parameters, and geometrically realistic features improve simulation fidelity.