# Planning Reivew

## Imitation Learning and RL
[Imitation learning](https://arxiv.org/pdf/2210.07729.pdf)
## Tesla
[Old 2019: imitation learning for planning and control](https://medium.com/@strangecosmos/a-theory-about-teslas-approach-to-imitation-learning-33b9e2c6d4d7)
- [Another blog](https://towardsdatascience.com/teslas-deep-learning-at-scale-7eed85b235d3)

2022 AI Day on Planning module
- 总体跟[TNT](https://arxiv.org/pdf/2008.08294.pdf)思路类似： [TNT作者解读](https://zhuanlan.zhihu.com/p/574101955)
- [更详细中文解释](https://zhuanlan.zhihu.com/p/570431956)
- [English version but oversimplified](https://driveteslacanada.ca/news/ai-day-2022-fsd-simplified/)
## Other
TNT作者新的simulator工作
- [InterSim](https://arxiv.org/pdf/2210.14413.pdf)
- [InterSim Github](https://github.com/Tsinghua-MARS-Lab/InterSim)
- The best use case for InterSim Beta is to test your planning system before deployment extensively. If you want to test your planner with a large real-world dataset, you should consider InterSim. InterSim models real-world drivers' behaviors against new conflicts with relation prediction models. InterSim now supports both **Waymo Open Motion Dataset** and the **NuPlan dataset**.
## Conclusion
You can brief review imitation learning (old methods) and RL-based approaches (impractical in real world). Generally the planning methods used by Tesla is quite similar to TNT. So you can talk more on the TNT paper for algorithm details (we reviewed this paper together before).