RoverSim is what you might an AI test-bed or playground and is the result of taking a homework assignment a few steps too far. Based off of a Scratch project, this is a complete rewrite in C#. Essentially, the goal is to write an AI that can collect, process, and finally transmit the most soil samples on the surface of Mars before it runs out of moves or power. The surface is represented by a grid with Smooth, Rough, and Impassable tiles.

More information about the project is available here, and a detailed description of each AI is available here.

All releases are available on GitHub. However, for the most up-to-date code, you should clone this repository and build from source.

Run either WinFormsClient.exe or RoverSim.AvaloniaHost.exe, depending on which client you're using. The Blazor Client isn't able to be run directly, but any HTTP server can host the site which can be viewed in any web browser supporting WebAssembly.

A typical AI will implement IAi through Simulate(IRoverStatusAccessor rover). Rover status is accessed through the IRoverStatusAccessor instance, which senses nearby terrain and various rover information. Actions are taken by yield returning a RoverAction - this will be applied by the caller and changes will be immediately visible through the status accessor. IAiFactory must also be implemented to create instances of the AI.

Once that's done, the factory needs to be added to the list of available AIs in each client. This is in WinFormsClient.Program.GetAIs, AvaloniaHost.Program.GetAvailableAis, and the constructor of AiProvider. For Blazor, the AI also needs to be added to the Renderer.razor component if you want it to be selectable.

Scratch AIs are restricted to a Scratch-like interface similar to the original in functionality. Specifically, ScratchRover does not provide a way to get the current coordinates - the AI must handle this by itself. Likewise, however, the AI may assume that it it will be running on the default simulation parameters, such as starting position and level dimensions. To write a Scratch AI, implement IScratchAi instead, which is provided with a ScratchRover instance. Then implement IAiFactory as usual, but instead wrapping the IScratchAi instance with an instance of ScratchAiWrapper and returning it instead.

As its name implies, Core is the heart of the project. Core defines the basic types, such as IRover and IAi, as well as simulation and terrain types. It also has classes to generate levels and to facilitate running millions of simulations using TPL dataflow. Everything else in the solution depends on Core.

Ais contains AIs written against the modern API. Currently, there is only LimitedStateAi, an AI that tries to make the most out of a limited amount of memory, as opposed to mapping out the entire level as it moves.

ScratchAis has all of the AIs originally written and destined for Scratch. This includes

• IntelligentRandom - always moves towards an adjacent smooth tile, or randomly otherwise.
• Mark I - similar to IntelligentRandom, but tries to get unstuck deterministically instead of randomly. The predecessor to LimitedStateAi.
• Mark II - uses Dijkstra's algorithm to always find the closest smooth tile.

WinFormsClient is a client built on Windows Forms (surprise!). With it, you can run millions of simulations and watch how the rover behaves on the worst one of the lot.

AvaloniaHost is a client built with the Avalonia UI framework. It aims to be a cross-platform client developed using modern UI techniques and patters. However, it is still a work in progress, with in depth statistics being the next goal.