VE-SIM: RL-Based Task Scheduling for Vehicular Edge Computing

A simulation framework for Reinforcement Learning (RL) based task scheduling in vehicular networks, integrating realistic traffic simulation with deep learning capabilities.

Key Features

• Integrated Simulation: Combines SimPy (discrete-event) with SUMO (traffic)
• RL-Ready Framework: PyTorch-based Agent and Environment
• Realistic Modeling: Car mobility + Task workloads with deadlines
• Flexible Scheduling: Scheduler supports heuristic and RL policies

Framework Components

1. Vehicles (Car class)

• Mobile compute units with:
  • Processing power, Task generation, Dynamic mobility via SUMO (TraCI)
• Handles task execution

2. Tasks (Task class)

• Computational workloads with:
  • Complexity, Deadline, Priority

3. Scheduler (Scheduler class)

• Core decision-making component:
  • Maintains system state (cars, tasks)
  • Implements policy matching
  • Supports:
    • Heuristics (EDF, priority-based)
    • RL policies

RL Integration

Component	Class	Functionality
RL Environment	TaskSchedulingEnv	Gymnasium interface for state/actions
DQN Agent	DQNAgent	Learns scheduling policy
Neural Network	DQN	Policy approximation
Experience Replay	ReplayBuffer	Training data storage

Dependencies

• SUMO (1.23.1)
• Python (3.11.10)
• for Python libraries see 'requirements.txt'

Getting Started

python3 main.py -cf config.ini -c RL-training --episodes 2000  --train