GoBigger Versions Save

New Features

• add practice which offer multiple gameplay modes for players to enjoy #56
• polish render’s visualization problem on leaderboard #56
• update numpy dependency version in setup.py #55
• update doc, config and citation #51 #52 #53

After GoBigger-Challenge-2021, we found some areas worth improving, including engine, space, and visualization tools. Here we introduce a new version of GoBigger to make it easier and more straightforward for users to develop their algorithms under multi-agent environments.

New Features

• Add owner for spore in overlap #37
• Add match_ratio in config to control density #38
• Allow balls to move over the border with the center and allow vision over border #39
• Engine update with more clear description #41
• Add different config #45 #51
• Add direction for each clone ball #46
• Add new replayer for .pb files #47

API change

• New ways to create environments! Please refer to env
• Simple action space! Please refer to action space
• Clear observation space! Please refer to observation space

The final version for challenge!

Bug fix

1. In gobigger/balls/clone_ball.py, self.radius + ball.radius may smaller than d.
2. In gobigger/players/human_player.py, self.get_clone_num() may equal 0.
3. Bug when the number of food balls is 0
4. flexible setting for render instead of fixed number

New Features

• Add bot-only mode with visual interface #22
• Add hyper action for RL #25 , please check Derived Environments and Action Spaces
• Add FAQ #23

Bug Fix

• Fix normalize bug #23
• Fix action_type = None with partial vision #24
• Update random engine to ensure reproducibility #25

API change

• We add seed parameter in Server for a quick way to set the seed.

New Features

• Add serveral ways to reload game for server #17

Bug Fix

• Fix overlap with speed
• Force player_name & team_name to be int
• Fix bug when vector.length() is 0 #20
• Use lower version numpy for python3.6

API change

• Change overlap in observation to reduce the memory #14

Enhancement

• Add filters before EnvRender.get_overlap() to reduce the num of balls to detect #15
• Decrease state_tick_per_second when doing training, because we don't need so many state tick to perform the game well #15
• Optimize PrecisionCollisionDetection in utils/collision_detection.py to get faster #15
• Update related doc #15

New Features

• Add custom initialization method #1
• Add speed & full vision in obs #7
• Add Human vs. Bot mode in interaction #7
• Add new bot of different levels #9

Bug Fix

• Fix server logit on spore ball #4
• Add bin in setup.py for quick call #6
• Engine: decouple direction for action and moving #7

API change

• Expanded action space! Please refer to action space

Enhancement

• Improve test coverage #5
• Add deep merge config #7
• Update default vel & vel_max to make moving faster #7
• Modified acceleration definition method #7
• Update interaction & action space & obs space #7
• Add advanced doc #10

Features

• Complete engine implementation in server.
• Accurate collision detection on different balls in the games.
• Standard format as gym.Env for RL environment.
• Designed action space and observation space for learning policy easily.
• Simple entry for users to interact with the games.