​

4 min presentation in DD2438

Group 4:19

Xiaolei Liu and Pedro Oliveira Mews

https://docs.google.com/presentation/d/13MBP8BQckle7YSDe952_C5_8TS2o7lFlJaGeovHvciU/edit?usp=sharing

Overview of our plan

• Familiarize with the new game engine
• Documentation, Quickstart and API Reference
• Search for initial references and ideas:
• Tree search with adaptive play out selection
• RL
• …

Adversarial N-player Search using Locality[1]

• Tree search algorithm using
• Alpha-beta pruning, when two agents involved
• Maxⁿ search, when more than two agents involved
• Agents involved are determined by locality:
• Only consider close agents
• Mask agents that are too far to directly interact with our agent
• This may lead to inaccuracies ->
• Heuristic to evaluate game state:
• Board control using Flood Fill
• More…

[1] Schier, Maximilian Benedikt; Wüstenbecker, Niclas (2019): Adversarial N-player Search using Locality for the Game of Battlesnake. SKILL 2019 - Studierendenkonferenz Informatik. ISBN: 978-3-88579-449-3. pp. 109-120.

Reinforcement Learning

• Maybe use this as reference https://aws.amazon.com/pt/blogs/machine-learning/building-an-ai-powered-battlesnake-with-reinforcement-learning-on-amazon-sagemaker/
• Gym_battlesnake, designed by Arthur Firmino,is devoted for RL in battlesnake game, which could be useful.
• Unlike last project, RL seems to be promising in this 2D small map

Progress Status Week 19

• Comment to customer paying 250 000kr for the report:
• “For the first week, we focused on searching for initial references, which explains the low progress. We will start the implementation and plan to have a working solution by next week.”
• Planned Time spent: 20%
• (Out of the combined 120h)
• Actual Time spent: 10%
• Out of the combined 120h
• Actual Progress: 8%
• (estimate progress towards completing assignment)
• Risk of not completing assignment: 3%