Advanced Robotics and Computer Vision: SROBO - An intelligent mobile ground robot proficient at autonomously navigating unstructured terrains using image segmentation and object classification

Proposed supervisory team

Dr Shabnam Sadeghi Esfahlani

Prof Hassan Shirvani

Alireza Sanaei

Theme

AI, Robotics and Automation. Impact themes: Health, Performance, and Wellbeing Sustainable Futures.

Summary of the research project

Research Topic: Enhancing mobile robot autonomy: exploring advanced algorithms for unstructured terrain navigation using deep learning.

Rationale: In the contemporary age of robotics, navigating unstructured terrains remains a significant challenge. While robots perform well in structured environments with clear paths and markers, terrains that are uneven, unpredictable, or filled with obstacles pose numerous complications. Leveraging the capabilities of advanced machine learning, particularly deep learning, offers a promising solution to this enduring challenge. Through the development and optimization of SROBO, this research seeks to push the boundaries of what is currently possible in robotic navigation, offering more versatile and adaptable robot deployment in real-world scenarios.

Objectives:

1. Develop a deep learning architecture tailored for real-time processing and optimized for mobile robotic platforms.
2. Investigate image segmentation techniques suitable for identifying varying terrains and their respective challenges.
3. Explore object classification methods to discern obstacles and navigate around or interact with them intelligently.
4. Benchmark and compare the performance of SROBO against traditional navigation algorithms in diverse terrains.
5. Assess the robustness of the developed algorithms against different lighting conditions, weather scenarios, and other environmental factors.

Methodology:

1. Dataset Compilation and Augmentation: Use a combination of available datasets for terrain and obstacle recognition and develop a custom dataset by operating SROBO in various terrains and conditions.
2. Deep Learning Model Development: Design and implement convolutional neural networks (CNNs) and potentially recurrent neural networks (RNNs) for sequential decision-making in navigation.
3. Image Segmentation: Employ semantic segmentation techniques to divide an image into segments, aiding in identifying paths, obstacles, and potential danger zones.
4. Object Classification: Utilize state-of-the-art classification models to identify and categorize obstacles, helping the robot decide on interaction strategies.
5. Real-time Integration and Testing: Integrate the developed models into SROBO's onboard computer and conduct real-world navigation tests in various terrains.
6. Performance Assessment: Deploy traditional navigation algorithms on SROBO as a comparison metric to evaluate the efficiency, accuracy, and reliability of the newly developed deep learning-based algorithms.

Expected Outcomes:

1. A refined version of SROBO with enhanced capabilities in autonomously navigating unstructured terrains.
2. A comprehensive understanding of the advantages and limitations of using deep learning techniques for robotic navigation in challenging environments.
3. A set of best practices and guidelines for deploying machine learning in mobile robots, particularly those operating in unpredictable terrains.

Contribution to the Field: This research aims to contribute significantly to the field of mobile robotics, offering a fresh perspective on navigating challenging terrains. The lessons learned from enhancing SROBO could potentially be applied to other robotic platforms, paving the way for more adaptable, intelligent machines in various applications, from exploration to disaster response.

Where you'll study

Chelmsford

Funding

This project is self-funded with the aim of completing in 3 years. Details of studentships for which funding is available are selected by a competitive process and are advertised on our jobs website as they become available.

Next steps

If you wish to be considered for this project, we strongly advise you contact the proposed supervisory team. You will also need to formally apply for our Engineering and the Built Environment PhD. In the section of the application form entitled 'Outline research proposal', please quote the above title and include a research proposal.