Update Webpage/Github Page/Documentation

[QuanyiLi]

• [x] Rebuild the website with videos
• [ ] Rebuild GitHub page with gifs/figs to show features
• [ ] Add relevant papers to doc/webpage. Give instructions about how to add paper to our list!
• [x] Use Grammarly to check!

Doc Sections:

• [x] Filter branch before merging
• [ ] Migrate Doc to another repo, it needed!
• [ ] It is possible to do 3Drendering on readthedoc server with xvfb

Installation:

• [x] manually pull asset, if can not find asset
• [x] cuda install

RL Training with MetaDrive:

• [x] Check existing content
• [x] Common config
• [x] 4 existing envs
• [x] specific env config
• [x] parallel: EnvPool/ray
• [x] Ray update to 2.0
• [x] StableBaseline
• [x] Add scenario environments
• [x] remove all gym.make/gym.register in doc and code!
• [x] Add horizon cfg to termination!

Concepts and customization:

• [x] General concepts: What is simulation?
• [x] Concrete System Design, Overview
• [x] Manager: a "mini" engine, a set of rules/why manager
• [x] config system. engine_config/env_config/access config everywhere and reduce the use of function parameters
• [x] tutorial for building a new environment (env.config point to engine.config)
• [x] How to add/access policy

Engine:

• [ ] Engine Lazy init
• [ ] BufferViewers
• [ ] coordinate system. Check all model directions!
• [x] Simulation time/repreat/physics step size
• [ ] Two physics world
• [ ] BaseRigid body

Objects:

• [ ] Object
• [x] Spawn Objects with the manager.
• [ ] Vehicle, Update vehicle config
• [ ] Map. BIG/blocks, self-define map, socket. Real-world maps. Connectivity, map bfs (make edge net bfs more efficient by filtering out nodes already in search list),
• [ ] Other objects
• [ ] TrafficLight, draw_lines, with SUMO example

Render:

• [x] top-down viewer, auto-scale, film-size, target_vehicle_heading_up, save map, coordinates and so on
• [ ] 3D viewer, Click to show lane and position
• [ ] Rotate camera is not compatible with CUDA!
• [ ] vglrun might be required for Headless running

Sensors:

• [ ] How to add/access sensors
• [ ] Lidar
• [ ] Camera (request main camera)
• [ ] Use one camera to capture multiple pics

GUI:

• [ ] panel
• [ ] Main Camera/First view cam
• [ ] Show sensor
• [ ] Show arrow

Others:

• [x] ScenarioNet: description, remind the user to set the first frame position to map center
• [ ] record/replay
• [ ] AgentManager, where and how to use agent config, change agent configs like steering_range via agent_configs.
• [x] developing new environments from scratch, add classes/managers step by step. See: environment/customization
• [ ] remind user to set map to center for real-data
• [ ] Manager information collection system

Debug:

• [x] debug tools: add_line/add_sphere/...
• [x] logging
• [x] p3d performance
• [x] debug_static_world
• [x] debug_physics_world
• [x] show_coordinates

Ecosystem:

• [x] ROS bridge
• [x] Sumo (TODO)
• [x] ScenarioNet
• [x] OpenDrive
• [x] DI-Drive

Others

• [ ] Update issue and PR template. A good reference is envpool
• [x] fix top-down line color
• [x] Test if horizon and max_step_per_agent still work.
• [x] Test if target_vehicle_config will overwrite vehicle_config
• [x] People don't need to get "vehicle_config" and update it before creating vehicles
• [ ] How to run test locally for developers

To clean:

• [ ] ray training
• [ ] generate_video

[QuanyiLi]

@pengzhenghao @BoSmallEar @WeizhenWang-1210 @pimpale

I plan to finish the new documentation before the end of this year. It will cover the content I mentioned above. Do you have any suggestions?

[CarlDegio]

1. the section "Read Data from Dataset" should be updated. This part is implemented through Scenarionet.
2. It is recommended to give the calculation process of vehicle dynamics simulation. Convenient for users to develop speed or trajectory control. Also for 'random_agent_model' in config.
3. DI-drive now give an exmaple about metadrive reinforcement learning. Can be added to the ecosystem.

[CarlDegio]

4. Introduction of expert policy (expert_takeover) can be added. And how to train a policy like this.