Automated Planning for Healthcare Logistics

🔍 The Challenge

Healthcare logistics in modern hospitals requires precise coordination of autonomous systems for efficient patient care and resource management. This project tackles the complexity of multi-agent robotic coordination in healthcare environments, where specialized robots must deliver medical supplies to patients while managing transportation logistics under realistic constraints.

Our comprehensive study demonstrates the evolution from classical STRIPS-based planning to sophisticated real-world implementations, providing insights into planning methodology trade-offs, scalability considerations, and practical deployment challenges through a distributed ROS2 architecture.

🧠 Problem Variants & Evolution

Our healthcare logistics domain encompasses five distinct problem variants, each building upon the previous to demonstrate the evolution of planning complexity:

Problem 1: Classical Planning (STRIPS)

Technologies: PDDL, Fast Downward, Fast Forward
Focus: Basic domain modeling with robot-box and robot-patient agents
Features: Fundamental actions for movement, supply management, patient transport

Problem 2: Enhanced with Carriers

Technologies: Extended PDDL with capacity constraints
Focus: Resource management and optimization
Features: Carrier objects with capacity constraints, load/unload actions

Problem 3: Hierarchical Planning (HTN)

Technologies: HDDL, PANDA planner
Focus: Task-oriented decomposition using Hierarchical Task Networks
Features: Abstract vs. primitive tasks, method definitions, structured problem-solving

Problem 4: Temporal Planning

Technologies: Temporal PDDL, OPTIC, TFD planners
Focus: Realistic timing requirements and concurrent execution
Features: Durative actions, temporal constraints, scheduling

Problem 5: ROS2 Implementation

Technologies: ROS2, PlanSys2 framework
Focus: Real-world distributed planning system
Features: Distributed architecture, action execution nodes, system orchestration

🛠 Methodology & Tools

Our multi-paradigm approach leverages different planning technologies:

Classical Planning Stack:

✅ PDDL for domain and problem specification
🔄 Fast Downward for optimal planning
⚡ Fast Forward for satisficing planning

Advanced Planning Technologies:

🏗️ HDDL & PANDA for hierarchical task decomposition
⏰ Temporal PDDL & OPTIC/TFD for time-aware planning
🤖 ROS2 & PlanSys2 for distributed real-time execution

Experimental Pipeline:

Domain Modeling - Healthcare logistics formalization
Scalability Analysis - Performance across problem dimensions
Comparative Evaluation - Methodology trade-offs
Real-World Deployment - ROS2 distributed implementation

🎬 ROS2 Implementation Demo

Real-world distributed planning system demonstration using ROS2 and PlanSys2 framework

🏥 Healthcare Domain Design

Our realistic healthcare scenario includes:

Robotic Agents:

Robot-box agents (logistics): Fill boxes with medical content, transport supplies
Robot-patient agents (care): Transport patients between treatment units

Environment:

Central warehouse for medical supply storage
Patient entrance and multiple treatment units
Interconnected locations with navigation constraints

Logistics Operations:

Medical supply delivery and distribution
Patient transportation and care coordination
Resource management with capacity constraints

📈 Key Performance Findings

Our comprehensive evaluation demonstrates significant insights across all planning paradigms:

✅ Classical planning provides optimal solutions but exhibits exponential complexity growth
📦 Carrier optimization reduces plan length by 25-40% compared to direct delivery
🏗️ Hierarchical approaches offer better scalability through task decomposition
⏰ Temporal planning enables 30-50% better resource utilization through parallelization
🤖 ROS2 implementation successfully maintains sub-second replanning capabilities

🔄 Real-World Validation:

Distributed architecture with PlanSys2 framework showing robust performance
Action execution nodes demonstrating seamless robotic command interfaces
System orchestration proving practical deployment viability in healthcare environments

🌐 Implementation & Architecture

Multi-Paradigm Development:

PDDL domain engineering for classical and temporal variants
HDDL hierarchical modeling for task decomposition
ROS2 node architecture for distributed execution
Comprehensive testing suite with scaled problem instances

Experimental Infrastructure:

Docker containerization for reproducible experiments
Performance benchmarking across different planners
Visualization tools for plan analysis and validation
Scaling test generation for systematic evaluation

🏥 Impact & Applications

This work demonstrates practical applications for:

Hospital automation and robotic logistics
Healthcare resource optimization and patient care coordination
Multi-agent planning in constrained environments
Real-time planning systems for dynamic environments

Future Extensions:

Integration with hospital information systems
Dynamic replanning for emergency scenarios
Learning-based plan optimization
Extended multi-robot coordination protocols

📈 Technical Contributions

Our study provides:

Comprehensive Methodology Comparison: Analysis across four major planning paradigms
Realistic Healthcare Modeling: Complex multi-agent logistics scenarios
Performance Analysis: Detailed scalability and efficiency studies
Real-World Implementation: Transition from offline to distributed real-time planning
Open-Source Framework: Complete implementation available for research community

BibTeX

@misc{decarlo2025planning,
  title={Automated Planning for Healthcare Logistics: From Classical to Real-World Implementation},
  author={De Carlo, Andrea},
  year={2025},
  institution={University of Trento},
  type={Course Project},
  note={Automated Planning and Temporal Planning}
}