demarsch / self-driving-car-sim Goto Github PK

I need to have a settings popup that will allow to configure the following parameters:

• Delta speed
• Delta angle
• Front FoV angle
• Front sensors number
• Rear FoV angle
• Rear sensors number

Add a band at the bottom of screen to visualize the training data. The band should consist of the number of frames that will fit the screen width. The band should be scrollable but not by-pixel, rather by-frame - this will allow to utilize virtualization.

Every frame should represent a picture of car, the data from its sensors and the action taken at that time point. User should be able to toggle between viewing training data and viewing last X (lets say 1000) actions taken.

When the frame viewer is in training data mode user should be able to delete every single frame or override the actions taken at that time point with the ability to batch apply these manual changes.

When the frame viewer is in 'last actions' mode, user should be able to pick specific frame or frame collections and add them to the training data.

I need to add a new mode where the car is actually displayed as a static one inside the center of the screen and the world should be scrolled in respective directions

The home page should have instructions on how to control the car and how to manage the obstacles

I need to have a way to download the training data and trained agent model and symmetric way to upload it. Need to consider the situations, with the custom number of front and rear sensors and custom field of view

• create a scene with physical bounds
• create a car a make it accelerate, decelerate and turn
• using ray-casting imitate a car FoV
• establish car communication with the environment in the form of action-state-reward
• design and implement one of the reinforcement learning techniques
• train the agent until it learns to avoid the obstacles
  ---------------------------------MVP----------------------------------------
• add extra visuals: car FoV, closest obstacle detector (line from car to obstacle - the closer the more thick and red it becomes, floating numbers - rewards and penalties)
• add configuration elements: level selector, obstacle adder, settings (e.g. penalty-reward ratio for specific actions)

I need to investigate the potential usage of matter.js timescale. This can improve the training as user won't record many similar states

Add ability to list the loaded agents in the toolbar to the left with options to import/export. Click on the agent should add the instance of this agent to the world and run the permanent auto-move (untill deleted manually). Click on the agent should remove it from the world