omritz/drl-crazyflie

How omritz/drl-crazyflie works

DRL-Crazyflie trains a DQN goal-seeking policy in an AirSim Gym environment, then can run a trained policy on a physical Crazyflie. Training turns simulator observations and rewards into replayed transitions and network updates; deployment instead loads a saved model and uses Crazyflie telemetry and motion commands. The map also retains the documented fixed-target-network and replay capabilities. Source excerpts support the core calls; checkpoint and model persistence are shown, but the saved artifacts are not inspected.

Main entry points: main_DRL.py Training runner · DQN_agent.py DQN agent · main_crazyflie.py Policy runner · AirGym.py Goal environment

Components 15

Training workflow

Training runner main_DRL.py
Training metrics main_DRL.py
Saved policy models
Weight checkpoints checkpoints

AirSim environment

Gym registration gym_airsim/__init__.py
Goal environment gym_airsim/envs/AirGym.py
AirSim controls gym_airsim/envs/myAirSimClient.py

DQN learning

DQN agent DQN_agent.py
Experience replay DQN_agent.py
Q networks DQN_agent.py

Crazyflie deployment

Policy runner main_crazyflie.py
Crazyflie controls myCrazyFlieClient.py

Other components

Project user
AirSim simulator
Crazyflie drone

Connections 19

Project user → Training runner: starts training README.md
Project user → Policy runner: runs policy README.md
Training runner → Gym registration: creates environment main_DRL.py
Gym registration → Goal environment: registers environment gym_airsim/__init__.py
Training runner → DQN agent: trains main_DRL.py
Training runner → Training metrics: records metrics main_DRL.py
Training runner → Saved policy: saves model
Training runner → Weight checkpoints: saves weights
Goal environment → AirSim controls: controls and observes gym_airsim/envs/AirGym.py
AirSim controls → AirSim simulator: sends commands gym_airsim/envs/myAirSimClient.py
AirSim simulator → AirSim controls: returns telemetry gym_airsim/envs/myAirSimClient.py
Training runner → Goal environment: steps environment main_DRL.py
DQN agent → Experience replay: stores and samples DQN_agent.py
DQN agent → Q networks: predicts and updates DQN_agent.py
DQN agent → Experience replay: samples transitions DQN_agent.py
Policy runner → Saved policy: loads model main_crazyflie.py
Policy runner → Crazyflie controls: requests observations and actions main_crazyflie.py
Crazyflie controls → Crazyflie drone: commands flight myCrazyFlieClient.py
Crazyflie drone → Crazyflie controls: provides telemetry myCrazyFlieClient.py

Drawn by GitDiagram on November 14, 2026 from the repository's file tree, README and source files. Read as Markdown ·