自动驾驶公司实习心得


自动驾驶公司实习心得

=======================

框架

开发工具

Docker

1.按照官网安装流程即可

https://docs.docker.com/engine/install/ubuntu/

2. 初步熟悉 docker

http://www.ruanyifeng.com/blog/2018/02/docker-tutorial.html

https://www.runoob.com/docker/docker-container-connection.html

gdb调试

Git

1. Git 开发流程简介

https://github.com/xirong/my-git/blob/master/git-workflow-tutorial.md#%E4%B8%89%E4%BC%81%E4%B8%9A%E6%97%A5%E5%B8%B8%E5%BC%80%E5%8F%91%E6%A8%A1%E5%BC%8F%E6%8E%A2%E7%B4%A2

2. 廖雪峰git教程

https://www.liaoxuefeng.com/wiki/896043488029600

3. 深入 Git 原理

https://git-scm.com/book/zh/v2

C++相关

算法相关

1. Reinforcement learning

快速review RL 常见算法: https://github.com/datawhalechina/easy-rl

  • MDP
    • Good explains: https://gaoyichao.com/Xiaotu/?book=probabilistic_robotics&title=pr_chapter14
  • POMDP ( https://arxiv.org/pdf/1109.2135.pdf)
    • Good explian: https://gaoyichao.com/Xiaotu/?book=probabilistic_robotics&title=pr_chapter15 
    • Offline
      • 基于点的值迭代
        • PBVI
          • 知乎简单讲解: https://zhuanlan.zhihu.com/p/168779238
          • 实现例子: https://zhuanlan.zhihu.com/p/272867881
          • (原论文: http://www.cs.cmu.edu/~ggordon/jpineau-ggordon-thrun.ijcai03.pdf)
    • Online
      • POMCP
        • 原论文: https://dspace.mit.edu/bitstream/handle/1721.1/100395/Silver_Monte-carlo.pdf?sequence=1&isAllowed=y
      • DESPOT
        • 原论文: https://proceedings.neurips.cc/paper/2013/file/c2aee86157b4a40b78132f1e71a9e6f1-Paper.pdf
  • 对 机器人规划控制的不确定性 的扩展
    • 概率机器人学
      • 中文版: https://gaoyichao.com/Xiaotu//resource/refs/PR.MIT.zh.pdf
      • 英文版: https://gaoyichao.com/Xiaotu//resource/refs/PR.manuscript.pdf

2. Imitation Learning

  • 简单入门教程  https://www.lamda.nju.edu.cn/xut/Imitation_Learning.pdf

3. 博弈论在自动驾驶交互的应用

VectorNet (https://blog.waymo.com/2020/05/vectornet.html)

  • 图神经网络(GNN) (DeepGraphLibrary库)
    • GAT (Graph Attention Network https://arxiv.org/abs/1710.10903)
    • GCN(Graph Convolutional Network https://arxiv.org/abs/1609.02907)
  • MLP
  • 自注意力机制 

LSTM 

  • Good explanation: https://colah.github.io/posts/2015-08-Understanding-LSTMs/

关于 MIO(Most Important Pbject) 的定义的论文→ https://hal.archives-ouvertes.fr/hal-01703415/document