AnimalAI - v4.1.0
Introducing Animal-AI v4.1.0!
NOTE: Make sure you change your 't' and 'pass_mark' parameters to 'timeLimit' and 'passMark' in the yaml files for this build.
So what's new?
New Feature/Gameobject: DataZones
Introducing a new game object called Datazone. The primary use case for this new game object is for data collection upon being entered by the agent or player. The data collected will be expanded in future iterations. In future versions, the Datazone is also to be used as a trigger to all sorts of arena-specific functionalities, such as spawning/de-spawning objects, opening doors and even handling multiple arenas in a single episode. Please check the example yaml file specified to get familiar with the syntax.
New Feature: LogDataToCSV
Introducing LogDataToCSV, an automatic data logging system that activates during both play and train modes. This system meticulously logs data about the environment, as well as the agent’s observations and actions, into a well-organized CSV file located in the root folder of AAI, specifically under ‘ObservationLogs’.
Key Features:
- Automatic Data Logging: Seamlessly logs and cleans data into a CSV file within the AAI root folder.
- Comprehensive Data Capture: Extensive data is recorded, as illustrated in the example CSV file provided.
- Mode Compatibility: Functions in both play and train modes, with optional feedback on training impact.
- User-Friendly: Designed to minimize disruptions to the user experience and workflow.
Feedback Request:
Your feedback is highly valued on the following aspects:
- Play Mode Functionality: Confirm that the application launches and closes as expected, without any performance issues such as crashes or freezes.
- Train Mode Stability: Ensure that the training process runs smoothly, without any unexpected freezes or crashes, and that the data is being recorded simultaneously (can be confirmed by checking the .csv, looking for any inconsistencies or gaps in rows).
- CSV File Generation: Verify that a CSV file is generated for each environment launch, named in the “Observations_date+time” format, applicable to both play and train modes.
- CSV File Readability: Assess the readability and clarity of the CSV file, focusing on the ease of understanding the captured data.
- Data Feedback: Provide any suggestions for additional columns or improvements in the data captured.
Thank you for your valuable input!
Updates:
(any updates/news to the experimental build will be displayed here)
- A background error was identified during testing where during training mode, AAI would freeze/crash upon launch. Fixed and uploaded new working build.
- Fixed bug where agent health was not updating during training mode. Uploaded new working build.
- (09.08) Added working build for Linux.
- (12.08) Fixed a bug where episodes were not sequentially incrementing (1,2,3,..,n) but were instead doubling randomly. Fixed and uploaded working build.
- (19.08) Experimental/pre-release status revoked and now listed as latest release. Uploaded final build version.
- 29.08) Re-uploaded new build to fix broken unity game objects Bounce and Grow/Ripen balls (due to erratic behaviour).
For a complete look, see CHANGELOG.md.
Getting Started:
Check our Installation Guide (here) and/or our Launching Animal-AI Guide (here) for a more detailed look.
But in brief:
- Download Animal-AI OS build for your system from below
- Extract files anywhere you'd like on your system (we use Winrar to package application contents)
- Run .exe/.app file (depending on your system) in the extracted folder of Animal-AI
N.B: If you are using Linux, you may need to make the .exe file executable. Simply run this command in your terminal of choice: chmod +x env/AnimalAI.x86_64. If you are using macOS, you may also need to run this command: chmod -R 777 [applicationName].app in your macOS terminal to unlock permissions for Animal-AI.
What's Changed
- Add documentation for multi-arena episodes by @benaslater in #51
New Contributors
- @benaslater made their first contribution in #51
Full Changelog: v4.0.0...v4.1.0.exp
