# Drone_learning

**Repository Path**: Joseph096/drone_learning

## Basic Information

- **Project Name**: Drone_learning
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 1
- **Forks**: 0
- **Created**: 2025-12-02
- **Last Updated**: 2026-01-30

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# Vision-based Agile Flight Training Code
进行一步修改
## Overview


[Project webpage](https://henryhuyu.github.io/DiffPhysDrone_Web/) is alive.

## Quick Demos
![alt text](image.png)

## Environment Setup
### Python Environment

The code is tested with the following environment:

- **PyTorch**: 2.2.2
- **Python**: 3.11
- **CUDA**: 11.8

The code should be compatible with other PyTorch and CUDA versions.

### Build CUDA Ops

To build the CUDA operations, run the following command:

```bash
#
pip install -e src
#其含义是将当前目录下的src文件夹里面的python项目，以可编辑模式安装到我的Python环境中
#正常 pip install 只会复制一份到site-package中，如果你修改了代码那么久必须重新pip-install 才会生效，但是一旦使用-e 那么随意修改src中的python文件，修改文件之后则立即生效，不需要重新安装

#现代项目越来越推荐 src 布局（把包代码放在 src/xxx 里），所以你现在看到 pip install -e src 的频率其实比以前更高了。
```

## Training

To start the training process, use the following command:

```bash
# For multi-agemt
python main_cuda.py $(cat configs/multi_agent.args)
# For single-agemt
python main_cuda.py $(cat configs/single_agent.args)
```

## Evaluation
You need to download the simulation validation code from the GitHub release page.
To evaluate the trained model in multi-agent settings, use the following command to launch the simulator:
```bash
cd <path to multi agent code supplementary>
./LinuxNoEditor/Blocks.sh -ResX=896 -ResY=504 -windowed -WinX=512 -WinY=304 -settings=$PWD/settings.json
```

## Citation
If using this repository, please cite our work
```
@article{zhang2025learning,
  title={Learning vision-based agile flight via differentiable physics},
  author={Zhang, Yuang and Hu, Yu and Song, Yunlong and Zou, Danping and Lin, Weiyao},
  journal={Nature Machine Intelligence},
  pages={1--13},
  year={2025},
  publisher={Nature Publishing Group}
}
```
Then, run the following command to evaluate the trained model:
```bash
python eval.py --resume <path to checkpoint> --target_speed 2.5
```