AIGP — Submission Prep

Quick Commands

python submit_check.py # validate all modules

python submit_check.py package # create submission.zip

python test_race_standalone.py # test race with dashboard

python rl_train.py train # train RL controller

Core Modules

✓race_pipeline.pyorchestrator + state machine

✓vision_pipeline.pycolor + YOLO + segmentation

✓mavsdk_bridge.pyMAVLink + 5 control modes

✓drone_mpc_foundation.pyMPC + pursuit controller

✓camera_adapter.pyGazebo/video/synthetic

✓race_config.pycentralized config + YAML

✓race_logger.pyJSONL logging

✓sim_drone.py6DOF physics

AI Modules

✓gate_segmentation.pyU-Net 3.7M params

✓rl_controller.pyPPO gym env + ONNX

✓rl_train.pySB3 training script

✓fpv_renderer.pyFPV for scenarios

iYOLO modeltrain with yolo-train.py

iRL policytrain with rl_train.py

iSeg modeltrain with gate_segmentation.py

Pipeline Architecture

Camera → Vision Pipeline → Gate Tracker → State Machine → Controller → MAVLink
          ↓                    ↓               ↓                ↓
   ColorDetector         EMA Smooth     SEEK/APPROACH      Pursuit (PID)
   YOLODetector          PnP Depth      TRANSIT/RECOVER    MPC (CasADi)
   SegDetector           Staleness      EMERGENCY          Neural (RL/ONNX)
                                        FINISHED

Competition Specs (VADR-TS-001)

Protocol	MAVLink v2 over UDP via MAVSDK
Control	SET_ATTITUDE_TARGET / SET_POSITION_TARGET_LOCAL_NED
Physics	120 Hz
Commands	50–120 Hz
Max time	8 minutes
Camera	Single FPV, ~12MP wide-angle
Compute	~100 TOPS (Jetson Orin NX)
No GPS	Local NED only
Round 1	Desaturated, highlighted gates
Round 2	3D-scanned photorealistic

Config Warnings

!hover_thrust = 0.50calibrate from sim

!gate_width = 2.0mupdate from spec

!camera FOV = 90°×60°measure from DCL

!HSV presetstune on VQ1 footage

isim_urlchange to DCL endpoint

Submission Package (48KB)

race_pipeline.py	Race orchestrator — main entry point
vision_pipeline.py	Gate detection + PnP depth
mavsdk_bridge.py	MAVLink communication
drone_mpc_foundation.py	Controllers + MPC
camera_adapter.py	Camera source adapters
race_config.py	Configuration
race_logger.py	Logging
+ AI models	gate_detector.onnx, policy.onnx, gate_seg.onnx

python submit_check.py package

SUBMISSION PREP

AI GRAND PRIX — PRE-FLIGHT VALIDATION