University of Toronto - Fall
2016
Department of Computer Science
CSC 2542:
Topics in KR&R: Algorithms for Sequential Decision Making
Computational Resources
Back to the main page
Most projects require code (implementation of algorithms) as well as data
to test your work. The latter may be in the form of benchmark problems,
testbeds or simulators.
This page contains a list of resources that may be of help to you in the
planning or realization of your projects. We will be updating this page
as we find more resources. We welcome suggestions for additions.
Also, if you end up using any of these resources, you may wish to create
a subgroup in Piazza to share your growing expertise. We also welcome
feedback on various resources.
Deterministic Planning
- http://planning.domains
A collection of tools for working with planning domains, including access
to a large suite of benchmark domains from over a decade of IPC (International
Planning Competition) domains, APIs for running different planner code, and
an editor for creating new domains in PDDL, the standard input format for (nonprobabilistic) planning problems.
- https://www.ida.liu.se/~TDDD48/labs/2016/planners.en.shtml
A great web page with an overview of different planners and instructions on how to run them, plus much more, by Jonas Kvarnstrom from Linkopings University.
Pathfinding Benchmarks
- http://movingai.com/benchmarks
A collection of benchmark problems for assessing pathfinding algorithms. These benchmarks include 2-D maps and mazes from video games as well as randomly
generated maps.
Frameworks for Heuristic Search
- PSVN: Please email profs for access.
A language for defining search problems, a compiler from the language to C code,
and search code to build pattern databases and run standard algorithms.
- HOG2: https://github.com/nathansttt/hog2
A search framework with a variety of algorithms and domains. The assignment code bases some of its code on HOG2.
- SBPL: http://www.sbpl.net
Code for search for robotics.
Motion Planning
Reinforcement Learning
Arcade Learning Environment (ALE)
- http://www.arcadelearningenvironment.org/
The Arcade Learning Environment (ALE) is a simple object-oriented framework that allows researchers and hobbyists to develop AI agents for Atari 2600 games. It is built on top of the Atari 2600 emulator Stella and separates the details of emulation from agent design.
Deep Reinforcement Learning
More Games Environments
- Mario: web --
http://www.marioai.org/
paper --
http://julian.togelius.com/Karakovskiy2012The.pdf
This API for playing Mario provides arrays that describe whether terrain is walkable and whether there are enemies.
The paper describes different approaches (learning based and hand-crafted) that entered a Mario AI competition.
- Minecraft:
https://www.microsoft.com/en-us/research/project/project-malmo/
This is a very popular game nowadays and Microsoft released Project Malmo: a challenge to create AI agents using the Minecraft environment.
It is interesting, as the API has access to discrete actions, grid mode (i.e. full observability) and also screen mode (i.e. video rendering). RL people may say the grid mode is "cheating", though.
Back to the main page