本文转载自https://zhuanlan.zhihu.com/p/157867488
平台:Gazebo
1. gymfc
作者在无人机姿态控制上使用PPO训练取得了比PID更好的效果,并成功从虚拟环境迁移到了现实世界。
项目总结书
相关论文:Reinforcement Learning for UAV Attitude Control,Neuroflight: Next Generation Flight Control Firmware
Github项目地址:https://github.com/wil3/gymfc
平台:Unity
1. Flightmare
Github项目地址:https://github.com/uzh-rpg/flightmare
使用指南:https://github.com/uzh-rpg/flightmare/wiki/Basic-Usage-with-Python
相关论文:Flightmare: A Flexible Quadrotor Simulator
2. FlightGoggles
主页:https://flightgoggles.mit.edu/
Github项目地址:mit-fast/FlightGoggles
3. ml-drone-collection
无人机姿态控制、导航。无人机有三、四、六、八旋翼。
介绍视频:https://www.youtube.com/watch?v=MKDBcKNJVS4
Github项目地址:https://github.com/mbaske/ml-drone-collection
4. FixedWingUAV
用DQN训练无人机避障
Github项目地址:https://github.com/phachara-lao
平台:虚幻4
1. AirSim
微软开发的仿真环境,可以通过手动或者程序来驱动仿真环境中的无人机
项目指南:https://microsoft.github.io/AirSim/
相关论文:AirSim: High-Fidelity Visual and Physical Simulation for Autonomous Vehicles
Github项目地址:microsoft/AirSim
2. PEDRA
用于无人机导航,状态输入为视觉图像,动作空间是离散的。环境挺多的,分为室内环境和室外环境。有3种类型的无人机。
相关论文:Autonomous Navigation via Deep Reinforcement Learning for Resource Constraint Edge Nodes using Transfer Learning
Github项目地址:https://github.com/aqeelanwar/PEDRA
3. Air Learning
相关论文:Learning to Seek: Deep Reinforcement Learning for Phototaxis of a Nano Drone in an Obstacle Field,Air Learning: An AI Research Platform for Algorithm-Hardware Benchmarking of Autonomous Aerial Robots
Github项目地址:harvard-edge/AirLearning
4. gym-airsim
无人机对人的正面视角进行拍摄,动作空间是离散的。这个环境是使用微软的AirSim搭建的。
相关论文:Leveraging Deep Reinforcement Learning For Active Shooting Under Open-World Setting
Github项目地址:https://github.com/tzimasak/gym-airsim
平台:PyBullet
- gym-pybullet-drones
Github项目地址:https://github.com/JacopoPan/gym-pybullet-drones
其他
- TelloPy
Github项目地址:hanyazou/TelloPy