Atem

What’s New

New Features

1. Interpreter Class:
• Introduced a new Interpreter class for direct TFLite model inference.
• Added functionality to interpret task sequences and sensor data.
• Supports saving interpretation results to a JSON file for easy logging and debugging.
2. Improved API Usability:
• Enhanced the AdaptiveModel and ModelTrainer classes with better task prediction and sequence encoding.
• Updated task sequence padding logic for increased flexibility.
3. Increased Example Input Coverage:
• Added more robust examples for inputs to ensure clarity in documentation and usage.

Whats New?
- Added improved documentation for getting started with ATEM

Whats new?
Improved ModelTrainer:
- Added the ability to dynamically adjust the maximum task sequence length with a new set_max_length method.
- Enhanced error handling for better user feedback when tasks or sequences are invalid.
Simplified Documentation:
- Unified docs.md with README.md for streamlined package documentation and setup.
New Features:
- Added terminal output color coding using colorama for a more engaging user experience during training.
Bug Fixes:
- Resolved issues with task sequence encoding and padding that caused errors with negative index padding.
- Fixed incorrect mappings in the task encoder.


This version is fully backward-compatible with v1.2.4. Users are encouraged to upgrade for improved stability and features.

1. Adaptive Model Integration
- Introduced the AdaptiveModel class for real-time task prediction based on sensor data.
- Enhanced predict_next_task function for dynamic task scheduling using TensorFlow Lite models.
2. Improved Package Structure
- Refactored the project into a cleaner directory structure:
- Core functionalities under `atem_core/core`.
- Utility functions under `atem_core/utils`.
- Added proper `__init__.py` files to support package discovery and modular use.
3. Real-World Simulation
- Developed a real-time simulation workflow using mock sensor data to test adaptive task prediction.

The ATEM (Adaptive Task Execution Manager) is a Python-based machine learning framework designed for FTC robots to dynamically adapt their task execution strategies. It integrates a TensorFlow Lite model with A* pathfinding to optimize autonomous operations on the FTC field.

Features

AI-Powered Task Selection:
- Processes real-time sensor data (e.g., time, distance, gyro, battery).
- Predicts the next optimal task using a trained TensorFlow Lite model.
A Pathfinding for Navigation*:
- Robot moves autonomously on the FTC field, avoiding obstacles.
- Compatible with FTC wheel and arm control functions.
Modular Design:
- Python package structure allows easy integration into robotics projects.
- Java interoperability to run directly on FTC robot controllers.
Robust Training Pipeline:
- Train adaptive models to fine-tune task prioritization.
- Generate, interpret, and deploy tasks dynamically.