Quantum-Agents
Hybrid Quantum-Classical Reinforcement Learning Agents
This repository contains practical implementations of agents that use quantum computing for decision-making and learning, combining variational quantum circuits (VQCs) with classical reinforcement learning algorithms.
Overview
Modern quantum agents are hybrid systems: they use parametrized quantum circuits as function approximators (policies or value functions) trained with classical optimizers. This approach works on today’s NISQ devices and simulators.
What’s Inside
- Jupyter Notebooks: Step-by-step tutorials with working code
- Hybrid RL Examples: Quantum policies for CartPole, MountainCar, and custom environments
- Architecture Patterns: Encoder → VQC → Classical readout pipelines
- Backend Support: Local simulators + cloud hardware (IBM, Amazon Braket, IonQ)
Quick Start
Installation
pip install pennylane pennylane-qiskit torch gym matplotlib
Run Your First Quantum Agent
jupyter notebook quantum_rl_agent_tutorial.ipynb
Architecture
Environment Observation
↓
Classical Encoder (NN)
↓
Quantum Circuit (VQC) ← trainable parameters
↓
Classical Readout
↓
Action Selection
Key Components
- Environment: OpenAI Gym or custom optimization problems
- Encoder: Maps observations to quantum circuit inputs
- VQC: Parametrized quantum circuit (2-12 qubits)
- Optimizer: Adam, SPSA, or COBYLA for training
- Backend: Simulators (development) → Hardware (experiments)
Features
- ✅ REINFORCE policy gradient implementation
- ✅ PennyLane + PyTorch integration
- ✅ Shot-noise simulation for hardware validation
- ✅ Multiple backend support
- ✅ Visualization and metrics
- ✅ Ready for cloud quantum hardware
Notebooks
quantum_rl_agent_tutorial.ipynb– Complete walkthrough with CartPole- More coming soon…