RL Agent for Dynamic Vehicle Routing & Resource Allocation

Overview

This project uses Reinforcement Learning (PPO) to solve Dynamic Vehicle Routing and Resource Allocation problems under stochastic customer demand and strict delivery time-window constraints.

The system simulates real-world logistics operations where delivery vehicles must optimize routing decisions in real time while minimizing route duration, delivery delays, and total travel distance.

---

Tech Stack

• Python
• PyTorch
• Stable-Baselines3 (PPO)
• Gymnasium
• Google OR-Tools
• MLflow
• Docker
• NumPy
• Pandas
• Matplotlib

---

Features

• Custom Gymnasium routing environment
• PPO-based Reinforcement Learning agent
• Dynamic delivery routing optimization
• Stochastic customer demand simulation
• Delivery time-window constraints
• OR-Tools optimization benchmark
• Offline evaluation framework
• MLflow experiment tracking
• Dockerized training pipeline

---

Problem Statement

Traditional vehicle routing systems struggle in dynamic environments where customer requests continuously change.

This project explores how Reinforcement Learning can improve:

• Real-time dispatch decisions
• Resource allocation
• Route optimization
• On-time delivery performance
• Logistics efficiency

---

Reinforcement Learning Pipeline

Environment → PPO Agent → Actions → Rewards → Policy Optimization

---

Evaluation Metrics

The RL agent is evaluated using:

Metric	Description
Average Delivery Time	Mean route completion time
On-Time Delivery %	Deliveries completed within deadline
Total Distance	Overall route distance
Fuel Cost	Estimated operational cost

---

Benchmark Comparison

The RL agent is benchmarked against:

• Nearest Neighbor Heuristic
• Google OR-Tools MILP Solver

---

Results

• RL agent achieved lower average route duration
• Improved delivery efficiency
• Reduced total travel distance
• Better adaptability under stochastic demand

---

Repository Structure

RL-Agent-for-Dynamic-Vehicle-Routing-Resource-Allocation/
│
├── env/
├── agents/
├── baselines/
├── configs/
├── experiments/
├── docker/
├── notebooks/
├── results/
│
├── train.py
├── evaluate.py
├── requirements.txt
└── README.md

---

Future Improvements

• Multi-agent Reinforcement Learning
• Graph Neural Networks (GNNs)
• Real-time traffic simulation
• Cloud deployment
• FastAPI inference API
• Streamlit dashboard

---

Author

Vishal Singh Mourya

MBA Data Science | Reinforcement Learning | Optimization | Machine Learning