DI-drive Documentation

DI-drive

Decision Intelligence Platform for Autonomous Driving simulation.

Last updated on 2022.6.5


DI-drive is an open-sourece application platform under OpenDILab. DI-drive applies Deep Learning method to decision, planning and controlling task in Autonomous Driving simulation, with high ease of use and low difficulty to hands-on. DI-drive designes Casezoo to make simulation environment closer to real driving. DI-drive develops Deep Learning driving policy with Pytorch and DI-engine and supports Carla and MetaDrive simulator.

Main features for DI-drive

  • AD Simulation

    DI-drive provides unified and easily used interfaces to support all kinds of lightweight and complex driving simulators. The user only needs to customize the input and output of the driving policy and simply call these interfaces to complete the simualtion.

  • Simplified training process

    DI-drive provides a variety of modules and tools to ease the training and testing of driving policies, including data collection, training and evaluation for Imitation Learning, standard gym.Env instance for Reinforcement Learning, and simple usage of DI-engine. This can greatly reduce the difficulty of deploying driving policy for beginners.

  • Modular Design

    Autonomous driving tasks are decomposed into Policy and Environment. Users can customize and modify simulation and training settings by changing the configuration files, without diving deep into complex models and internal details of the simulator. DI-drive defines policy in a flexible, polymorphic and efficient way. It can adapt to all existing academic literatures and support complex training tasks across methods, models, datasets, and even across simulators.

  • CaseZoo sets

    DI-drive designs a Casezoo simulation set, by integrating existing Autonomous Driving evaluation indicators, scenarios and tools in both academia and industry. Casezoo combines data collected by real vehicles and Shanghai Lingang road license test Scenarios. Casezoo can realize scenario-based Autonomous Driving test, makes the simulation closer to real driving. Di-drive supports to test as well as train Renfircement Learning policy with Casezoo environment.


This is home page contains description of tutorials, features and API documentation for DI-drive. Feel free to check any part you need. It is recommended to first install DI-drive and have a quick try following provided quick start guidence.

Table of Contents