01. Wheelly - m-marini/wheellyj GitHub Wiki
Wheelly
Release 0.27.0
The application uses a neural network agent to control the robot.
Run
After installing the application:
bin\wheelly
Java application
The application can also be started with:
bin\runnit org.mmarini.wheelly.apps.Wheelly
Options
-h, --help Show help
-a, --alternate Alternate act/training mode
-c, --config CONFIG YAML configuration file
-i, --inference PATH Inference output path
-s, --silent Silent mode
-t, --localTime SEC Session duration in seconds
-v, --version Show version
-w, --windows Use multiple windows
The default session duration is 43200 seconds (12 hours).
Automated sessions
Workflow
flowchart TD
A[Configure] --> B[Run Wheelly]
B --> C[Generate CSV data]
C --> D[Generate report]
D --> E{.running exists?}
E -->|Yes| B
E -->|No| F[Stop]
Each session runs for:
172800 seconds = 48 hours
The loop continues while the .running file exists. Remove this file to stop after the current cycle.
Output
The main output directories are:
%KPIS%— KPI data%CSVS%— CSV data%REPORT%— reports
The previous contents of these directories are deleted before each new session.
Note: GNU Octave 7.2.0 must be installed in the configured path.