Multi-Robot Fleet Management System (FMS)

Overview

This project implements a Fleet Management System (FMS) in a 2D grid-based warehouse environment.
Multiple robots navigate, pick up boxes, deliver them to drop zones, avoid collisions, and manage battery levels.

---

What is FMS (Fleet Management System)?

A Fleet Management System is responsible for:

• Assigning tasks to robots
• Planning paths for navigation
• Avoiding collisions
• Managing resources like battery

In real-world warehouses (like Amazon), FMS ensures efficient and safe robot coordination.

---

Without RL (Current Implementation)

In this project, robots operate using:

🔹 Path Planning

• Grid-based movement (Up, Down, Left, Right)
• Local navigation using 5x5 observation window
• Distance-based movement towards target

🔹 Task Allocation

• If robot is not carrying → nearest box assigned
• If robot is carrying → nearest drop zone assigned

🔹 Collision Avoidance

• Swap detection (robots swapping positions blocked)
• Destination conflict handling (multiple robots same cell blocked)

🔹 Battery Handling (Basic)

• Battery decreases every step
• Charging stations restore battery
• No intelligent scheduling (rule-based)

👉 This is mostly rule-based + heuristic system

---

With RL (Future Scope)

If Reinforcement Learning is applied, system becomes much smarter:

🚀 Advanced Capabilities with RL

• Learn optimal task allocation
• Minimize total travel time
• Reduce collisions automatically
• Learn when to charge vs when to deliver
• Handle multi-robot coordination dynamically

🔋 Battery Optimization

• Decide best time to go to charger
• Avoid unnecessary charging
• Balance delivery vs charging

📉 Performance Improvements

• Fewer collisions
• Faster deliveries
• Better resource utilization

---

Environment Design

Grid Representation

• 0: Free Space
• 1: Obstacle
• 2: Charging Station
• 3: Box (Pickup)
• 4: Drop Location
• 5: Robot

---

Robot Capabilities

• Move in 4 directions + stay
• Pick and drop boxes
• Battery consumption per step
• Local perception (5x5 grid like LiDAR)
• Collision avoidance

---

Core Features

1. Multi-Robot Coordination

• Multiple robots operate simultaneously
• Collision handling:
  • Swap conflict prevention
  • Destination conflict resolution

2. Task Allocation (Rule-Based)

• Nearest box assigned if not carrying
• Nearest drop assigned if carrying

3. Battery Management

• Battery decreases each step
• Charging station restores battery
• Low battery threshold triggers charging (basic logic)

---

Reward System (for RL compatibility)

Positive Rewards

• +50 → Delivery completed
• +5 → Box picked
• +200 → All deliveries completed
• +1 → Charging when battery is low
• +distance improvement → moving closer to target

Negative Rewards

• -0.5 → Collision / invalid move
• -50 → Battery depletion

---

Observation Space

Each robot gets a 31-dimensional vector:

• 5x5 local grid (25 values)
• Battery level
• Carrying status
• Relative target position
• Relative charger position

---

Action Space

• 0: Up
• 1: Down
• 2: Left
• 3: Right
• 4: Stay

---

Difficulty Levels

Easy (easy_delivery)

• 1 box
• Low battery drain
• Simple navigation

Medium (multi_order)

• 4 boxes
• Moderate battery drain
• Requires coordination

Hard (hard_fleet)

• 8 boxes
• High battery drain
• Complex coordination + charging

---

Episode Termination

• All boxes delivered
• Battery reaches 0
• Max steps reached

---

Key Contributions

• Designed a custom multi-robot simulation environment
• Implemented collision-safe movement logic
• Built task allocation and navigation system
• Created RL-compatible observation + reward system

---

Conclusion

This project demonstrates how a Fleet Management System can be built using rule-based planning.

While current implementation uses heuristics (path planning + allocation),
it can be extended with Reinforcement Learning to achieve:

• Smarter decision making
• Better coordination
• Efficient energy usage
• Reduced collisions

👉 This makes it a strong foundation for real-world warehouse robotics systems.