Examples

Runnable examples on the framework, plus the Isaac Lab RL app, a separate uv project. Launch any with uv run -m nexus_uav.examples <name>, and --list shows them all. Each example is zero-arg: its configuration lives in the script. Each records an interactive Rerun .rrd of its flight and needs a CUDA device. A vehicle that authors RTX sensors renders them in the Kit render peer, which the example starts on its own.

• PID waypoint tour: the minimal device-native controller, a deterministic, differentiable Proportional Integral Derivative (PID) controller flying a waypoint square, fully CUDA-graph captured.
• Differentiable design optimization: back-propagate through a closed-loop flight to tune a PID and recover a mass.
• acados NMPC: a real-time Nonlinear Model Predictive Control (NMPC) controller flying a curving waypoint course nose-first along a min-snap reference.
• Sampling MPC slalom: a sampling-plus-gradient differentiable-simulation MPC flying an obstacle slalom.
• Astro Max RL on Isaac Lab: a swarm of parallel agents learning to hover with Proximal Policy Optimization (PPO), then the policy deployed on the core runtime.
• PX4 SITL flight through a conformed actuator: a PX4 Software In The Loop (SITL) flight through a conformed per-rotor actuator, with the real-time factor reported.