# Joint Research and Collaboration Proposal https://www.researchgate.net/profile/Md-Hasibuzzaman-8?ev=hdr_xpr…

By Md Hasibuzzaman

# Joint Research and Collaboration Proposal

https://www.researchgate.net/profile/Md-Hasibuzzaman-8?ev=hdr_xprf

https://scholar.google.com/citations?user=KRdmGLIAAAAJ&hl=en

## Topology-Aware Intelligent Navigation for Autonomous and Multi-Robot Systems

**Researcher:** Md. Hasibuzzaman

**Affiliation:** PhD Researcher, Department of Electrical Engineering and Computer Science (EECS), National Taipei University (NTPU), Taiwan

### Research Background

My research interests are centered on **autonomous robotics, intelligent navigation, dynamic path planning, computational geometry, skeletonization, graph-based representations, reinforcement learning, computer vision, and uncertainty-aware autonomous control**.

My previous research has investigated skeleton-based representations of free space, Enhanced Grassfire (EGF) and wave-propagation-based skeletonization, dynamic path planning, collision avoidance, and navigation in environments containing moving obstacles and targets. My more recent research direction extends toward **reinforcement learning, uncertainty-aware decision-making, and robust vision-based robotic control**.

These research directions provide a strong foundation for developing a collaborative research program that connects **geometric/topological knowledge with modern learning-based autonomous systems**.

---

## Proposed Collaborative Research

I would like to propose a collaborative research project focusing on:

### **Topology-Aware Reinforcement Learning for Safe Autonomous Navigation in Unknown and Dynamic Environments**

The central idea is to combine **topological/skeleton-based environmental representations with reinforcement learning**.

Traditional path-planning algorithms can exploit the geometric structure of an environment but may require frequent replanning when the environment changes. In contrast, reinforcement learning can learn adaptive behaviors but may suffer from unsafe exploration, poor generalization, and sensitivity to uncertainty.

The proposed research will investigate whether these limitations can be addressed by providing an RL agent with a **dynamically updated topological representation of the environment**.

---

## Core Concept

The proposed architecture will follow:

**Sensor Perception**

↓

**EGF / Wave-Propagation Skeletonization**

↓

**Topological Graph Construction**

↓

**Uncertainty Estimation**

↓

**Topology-Aware Reinforcement Learning**

↓

**Safety-Aware Decision Making**

↓

**Autonomous Robot Control**

Instead of learning navigation directly from raw sensor observations, the agent will receive both conventional observations and structured information describing the topology of the navigable environment.

The topological representation can contain:

* navigable corridors;

* critical junctions;

* connectivity;

* obstacle clearance;

* path length;

* local geometric characteristics;

* dynamic obstacle information;

* environmental uncertainty.

This representation can then be processed using graph-based learning or graph neural networks.

---

## Main Research Question

The fundamental question of the proposed collaboration is:

> **Can dynamically evolving topological knowledge make reinforcement-learning-based autonomous navigation safer, more efficient, generalizable, and robust under environmental uncertainty?**

---

## Key Novelty

The proposed work is not intended to be a simple combination of skeletonization and reinforcement learning.

The primary scientific contribution will be the development of a framework in which:

> **Dynamic topological knowledge acts simultaneously as a structured state representation, learning prior, and safety constraint for reinforcement-learning-based navigation.**

This creates a bridge between **model-based geometric intelligence and data-driven learning-based intelligence**.

---

## Uncertainty-Aware Navigation

Real-world robots frequently operate under uncertain conditions caused by:

* sensor noise;

* incomplete observations;

* moving obstacles;

* localization errors;

* visual occlusion;

* temporary sensor failures;

* changing environmental topology.

The proposed framework will therefore explicitly estimate navigation uncertainty and adapt the control strategy accordingly.

Under low uncertainty, the learned policy can operate normally.

Under high uncertainty, the system can rely more strongly on a topology-based safe navigation policy.

This will provide a mechanism for **adaptive and safety-aware policy selection**.

---

## Multi-Robot Extension

A further research direction will extend the proposed framework to **multi-robot systems**.

Instead of sharing complete maps between robots, each robot can exchange compact topological information such as:

* important graph nodes;

* available corridors;

* obstacle locations;

* corridor occupancy;

* navigation priorities;

* uncertainty information.

This could enable efficient decentralized coordination while reducing communication overhead.

The resulting framework would therefore support:

**Topology + Learning + Uncertainty + Multi-Robot Coordination**

within a unified architecture.

---

## Research Methodology

The research can be developed through four major stages.

### Stage 1 — Topological Environment Representation

Develop dynamic EGF/skeleton-based representations and convert them into navigational graphs.

### Stage 2 — Topology-Aware Reinforcement Learning

Develop RL policies conditioned on both robot observations and topological graph representations.

### Stage 3 — Uncertainty and Safety

Introduce uncertainty estimation and a safety-aware mechanism for switching or blending between learned and model-based navigation policies.

### Stage 4 — Multi-Robot Navigation

Extend the framework to decentralized multi-robot navigation and investigate scalability and communication efficiency.

---

## Experimental Evaluation

The proposed framework will be compared against established approaches such as:

* A*;

* Dijkstra;

* D*;

* classical skeleton-based planners;

* PPO;

* SAC;

* TD3;

* conventional deep-RL navigation;

* topology-aware RL;

* the proposed uncertainty-aware framework.

Experiments will consider:

1. static environments;

2. unknown environments;

3. dynamic obstacles;

4. partially observed environments;

5. sensor uncertainty;

6. perception failures;

7. multi-robot environments.

Performance will be evaluated using:

* success rate;

* collision rate;

* path length;

* navigation time;

* minimum obstacle clearance;

* computational cost;

* robustness;

* generalization;

* recovery performance;

* communication overhead for multi-robot scenarios.

---

## Potential Collaborative Roles

My contribution can focus on the areas directly connected to my existing research expertise:

* skeletonization and EGF-based environment representation;

* dynamic path planning;

* topological graph construction;

* obstacle-aware navigation;

* reinforcement-learning-based navigation;

* uncertainty-aware decision-making;

* experimental implementation and evaluation.

The collaborating researcher/team could contribute complementary expertise in areas such as:

* multi-robot systems;

* graph neural networks;

* advanced reinforcement learning;

* robot hardware;

* distributed control;

* autonomous systems;

* theoretical analysis.

The exact division of responsibilities can be adapted according to the collaborator's research interests and laboratory capabilities.

---

## Expected Contributions

The proposed collaboration aims to produce:

1. **A novel topology-aware reinforcement-learning framework for autonomous navigation.**

2. **An uncertainty-aware navigation mechanism for robust operation under imperfect perception.**

3. **A safety-aware integration of learning-based and model-based navigation.**

4. **A decentralized topological representation for multi-robot coordination.**

5. **A comprehensive benchmark covering static, dynamic, uncertain, and multi-robot environments.**

6. **Open-source implementation and reproducible experiments.**

7. **High-quality journal/conference publications and potential future joint grant proposals.**

---

## Long-Term Research Vision

The broader objective is to develop autonomous robots that do not rely exclusively on either predefined mathematical models or end-to-end machine learning.

Instead, the robot should combine:

**Geometric Knowledge**

+

**Topological Reasoning**

+

**Learning-Based Intelligence**

+

**Uncertainty Awareness**

+

**Safe Control**

to achieve reliable autonomous behavior in complex real-world environments.

The proposed research therefore provides a natural progression of my existing research:

**Skeletonization → Topological Representation → Dynamic Path Planning → Reinforcement Learning → Uncertainty-Aware Navigation → Safe Autonomy → Multi-Robot Intelligence**

---

## Proposed Collaboration Statement

I am interested in establishing a research collaboration in this direction where our complementary expertise can be combined to develop a genuinely novel framework for intelligent autonomous navigation.

As a **PhD researcher in the Department of Electrical Engineering and Computer Science (EECS) at National Taipei University (NTPU)**, I am particularly interested in developing this work into a longer-term research collaboration involving joint publications, experimental validation, open-source research, and potentially future collaborative research funding.

### Collaboration Vision

> **To develop a new generation of autonomous robots that combine topological understanding with reinforcement learning and uncertainty-aware control to navigate safely, intelligently, and collaboratively in unknown and dynamic environments.**