Proactive Self-Adaptation in Autonomous Mobile Robots via Predictive Feasibility Reasoning

MSc Thesis - Mohamed Msallak - TU Delft - January 2025

Overview

Implementation of a proactive self-adaptation framework for energy-constrained autonomous mobile robots. The system extends ROSA with rolling-horizon battery prediction and integrates PlanSys2 for symbolic task planning, enabling anticipatory task-and-architecture co-adaptation.

Key features:

• Rolling-horizon energy prediction over upcoming navigation actions
• Proactive configuration switching and recharge planning
• Benchmark evaluation across procedurally generated environments
• Comparison with reactive threshold-based baseline

⚠️ Compatibility Notice: This system has been developed and tested exclusively on ROS 2 Humble. Compatibility with other ROS 2 distributions is not guaranteed.

Installation

Prerequisites

This project builds on the ROSA framework. Follow ROSA installation instructions:

• ROSA repository: https://github.com/kas-lab/rosa

Additional dependencies:

• ROS 2 Humble
• PlanSys2
• TypeDB 2.x
• Nav2 navigation stack

Build

mkdir -p ~/ros2_ws/src
cd ~/ros2_ws/src
git clone [this-repo]
cd ~/ros2_ws
colcon build
source install/setup.bash

Running the System

Prerequisites for All Runs

1. Start TypeDB Server (Terminal 1)

typedb server &

2. Start Battery Monitor (Terminal 2)

For benchmark mode:

ros2 run navigation_battery battery_monitor --ros-args -p benchmark_mode:=true

For single navigation runs:

ros2 run navigation_battery battery_monitor --ros-args -p benchmark_mode:=false

⚠️ Important: The battery monitor must be running before launching the navigation system.

---

Running the Benchmark

Proactive Adaptation (Terminal 3)

ros2 run navigation_planner benchmark_tracker

Reactive Adaptation Baseline (Terminal 3)

ros2 launch navigation_task_plan navigate_task_plan_reactive.launch.py \
  scenario_difficulty:=easy \
  benchmark_mode:=true \
  max_benchmark_runs:=100

Benchmark Parameters:

• scenario_difficulty:= - Map difficulty: easy, medium, or hard
• benchmark_mode:= - Set to true for automated benchmark runs
• max_benchmark_runs:= - Number of episodes to run
• enable_debug_in_benchmark:= - Set to true for debug output

Results

Benchmark results are saved to:

• benchmark_results/ - Metrics, success rates, configuration usage
• real_sim_results/ - Simulation validation data

---

Running Single Navigation Episodes

Useful for testing and debugging individual scenarios.

Single Run - Proactive Adaptation

Terminal 1: TypeDB

typedb server &

Terminal 2: Battery Monitor

ros2 run navigation_battery battery_monitor --ros-args -p benchmark_mode:=false

Terminal 3: Navigation (Proactive)

ros2 launch navigation_task_plan navigate_task_plan_horizon.launch.py

Single Run - Reactive Adaptation

Terminal 1: TypeDB

typedb server &

Terminal 2: Battery Monitor

ros2 run navigation_battery battery_monitor --ros-args -p benchmark_mode:=false

Terminal 3: Navigation (Reactive)

ros2 launch navigation_task_plan navigate_task_plan_reactive.launch.py \
  scenario_difficulty:=easy \
  benchmark_mode:=false \
  max_benchmark_runs:=1 \
  enable_debug_in_benchmark:=false

---

Repository Structure

├── navigation_kb/              # TypeDB knowledge base
│   ├── config/
│   │   ├── schema.tql         # KB schema definition
│   │   └── data.tql           # Static domain knowledge
│   └── launch/                # KB launch files
│
├── navigation_rosa/            # ROSA integration
│   ├── config/
│   │   ├── navigation_rosa.tql          # Proactive constraints
│   │   └── navigation_rosa_reactive.tql # Reactive constraints
│   └── launch/                # ROSA launch files
│
├── navigation_task_plan/      # PlanSys2 task planning
│   ├── pddl/
│   │   ├── domain_sas.pddl            # PDDL domain (proactive)
│   │   └── domain_sas_reactive.pddl   # PDDL domain (reactive)
│   ├── src/
│   │   ├── navigate_horizon.cpp       # Proactive controller
│   │   └── navigate_reactive.cpp      # Reactive controller
│   └── launch/
│       └── benchmark_with_map_regen.launch.py  # Main benchmark
│
├── navigation_planner/         # Benchmark & map generation
│   └── navigation_planner/
│       ├── benchmark_tracker.py       # Metrics collection
│       └── map_generator.py           # Procedural map generation
│
├── navigation_battery/         # Battery monitoring
│   └── navigation_battery/
│       └── battery_monitor.py         # Battery discharge model
│
├── benchmark_results/          # Benchmark output data
└── real_sim_results/           # Simulation validation data

Key Configuration Files

Proactive Thresholds

Located in navigation_rosa/config/navigation_rosa.tql

Reactive Thresholds

Located in navigation_rosa/config/navigation_rosa_reactive.tql

PDDL Domain

• Proactive: navigation_task_plan/pddl/domain_sas.pddl
• Reactive: navigation_task_plan/pddl/domain_sas_reactive.pddl

Map Difficulty Parameters

Configured in benchmark launch file parameters:

• easy: Dense connectivity, multiple charging stations
• medium: Moderate connectivity, limited chargers
• hard: Sparse connectivity, clustered chargers

---

Starting ROSA Independently

If you need to launch ROSA separately (not automatically started by navigation launch):

ROSA (Proactive)

typedb server &
ros2 launch navigation_rosa navigation_rosa.launch.py

ROSA (Reactive Baseline)

typedb server &
ros2 launch navigation_rosa navigation_rosa_reactive.launch.py

---

Troubleshooting

Battery Monitor Not Publishing

Check that battery monitor is running:

ros2 node list | grep battery

Check battery topic:

ros2 topic echo /battery_level

TypeDB Connection Issues

Verify TypeDB server is running:

ps aux | grep typedb

Restart if needed:

pkill typedb
typedb server &

Benchmark Not Starting

Ensure all prerequisites are running:

1. TypeDB server
2. Battery monitor in correct mode
3. Check launch file parameters

---

Repository Status

Note: This repository contains research code developed during an MSc thesis project. The implementation prioritizes experimental validation and rapid iteration over production-level software engineering. The defense-submission branch contains the working system used for thesis evaluation.

Thesis

Defense: January 27, 2025
Supervisors: Gustavo Rezende Silva, Carlos Hernández Corbato
Document: [Available after defense]

Citation

@mastersthesis{msallak2025proactive,
  title={Proactive Self-Adaptation in Autonomous Mobile Robots via Predictive Feasibility Reasoning},
  author={Msallak, Mohamed},
  year={2025},
  school={Delft University of Technology}
}

Contact

Mohamed Msallak - M.Msallak@student.tudelft.nl

License

See LICENSE file for details.