Open Projects!

The projects are all at present available for the following types: Master Thesis (MA), Bachelor Thesis (BA) and Semester Project (SA). 

The projects are supposed to be carried out in HKUST, ECE department. According to the latest top univertity ranking, HKUST is ranked 1st in Asian and its ECE department is ranked No. 14 worldwide, which has 13 IEEE fellows. The joint facilities there will provide premier environment for your research, experiments and technique supports.

Please contact Ming Liu for details. Further projects in related directions (e.g. Machine Learning, Computer Vision, Navigation system, Mobile robotics, Multi Robot System, GPGPU computing...) can be set up JUST FOR YOU, up on request.

If you are not in Hong Kong, grants to cover your return flight to Hong Kong and accommondation in HKUST is forseeable.


UST-01A Exploration, Mapping and Navigation using multi robot system (MA/BA)

This project is about the development of a full SLAM (Simultaneous Localization and Mapping) system using heterogeneous sensors and multiple robots. The aim is to develop advanced algorithms to efficiently implement exploration and SLAM for a large area using several mobile robots.

Preliminary requirements:

  1. Basic knowledge on SLAM or passion to learn it
  2. Programming background: prefer C/C++, python
  3. Probability theory or Graph theory or mathematical background

Core work packages:

  1. Learn ROS
  2. Development of the framework for multi robot cooperation
  3. Dataset gathering
  4. Algorithm designation for exploration task
  5. Documentation

Bonus work packages:

  1. Demonstration in real dynamic environment
  2. Paper writing for journals or top conferences

Please contact Ming Liu for details.

UST-01B Joint Resource/Task allocation Framework for multi robot system (MA/BA/SA)

UST01B
UST01B

This project is to develop a generic framework to jointly manage the resources (such as WiFi bandwidth and access to a centralized data center) and tasks (roles and targets for different robots) for a multi robot SLAM system. The aim is to develop advanced algorithms to realize optimal allocation of limited resources for the connected agents.

Preliminary requirements:

  1. Basic knowledge on optimization or game theory or passion to learn
  2. Programming background: prefer python, matlab

Core work packages:

  1. Setup a proper simulation environment for final validation
  2. Development of a joint framework for resource and task allocation
  3. Implementation and Validation
  4. Documentation

Bonus work packages:

  1. Paper writing for journals or top conferences

Please contact Ming Liu for details.

UST-02 Topological visual navigation in a dynamic environment (MA/SA/BA)

UST02
UST02

This project is a continuation of the paper "Visual Homing from Scale with an Uncalibrated Omnidirectional Camera, IEEE Transactions on Robotics, 2013". A topological visual navigation with loop-closure detection is to be developed within the project. As the only exterinsic sensor for the system, an omnidirectional camera is used to provide 360 degree field-of-view images for the system.

Preliminary requirements:

  1. Basic knowledge on visual keypoints (SIFT, SURF etc.) and matching
  2. Programming: c/c++

Core work packages:

  1. Learn ROS
  2. Get knowledge of existing algorithms
  3. Design and implementation of novel integrated system for topological visual navigation
  4. Documentation

Bonus work packages:

  1. Paper writing for journals or top conferences

Please contact Ming Liu for details.

UST-03 Dynamic object detection using omnidirectional camera / Kinect sensor (MA/BA)

This project aims at designing an algorithm to detect the dynamic objects in the environment and implementation on a mobile robot, using an omnidirectional camera or Kinect sensor. It's preferred to design the algorithm using learning techniques and feature based descriptors such as SIFT.

Preliminary requirements:

  1. Basic knowledge on computer vision or machine learning
  2. Programming: c/c++ or python

Core work packages:

  1. Learn ROS
  2. Designation of bayesian algorithm for dynamic object detection
  3. Implementation on real system
  4. Documentation

Bonus work packages:

  1. Implementation on mobile platforms
  2. Paper writing for journals or top conferences

Please contact Ming Liu for details.

UST-04 Stereo vision / kinect -based mobile manipulator (MA)

UST04
UST04

This project is to combine manipulator with a mobile robot, i.e. mobile manipulator system. The system uses stereo vision or kinect sensors to localize itself and perform manipulation operations efficiently. Realtime ability is also to be considered as the the key characteristic for the final system.

Preliminary requirements:

  1. Basic knowledge on computer vision or/and manipulator geometry
  2. Programming: c/c++, matlab

Core work packages:

  1. Learn ROS
  2. Designation and construction of the prototype system
  3. Algorithm designation for efficient mobile manipulating
  4. Implementation and validation
  5. Documentation

Bonus work packages:

  1. Fulfill real-time demonstration
  2. Paper writing for journals or top conferences

Please contact Ming Liu for details.

UST-05 Task allocation for multi robot system

UST05
UST05

The test system is a union of several small scale educational robots. The aim of the project is to coordinate multi robots to finish a unified complex task together, such as moving large objects jointly. 

Preliminary requirements:

  1. Programming: prefer python, C/C++

Core work packages:

  1. Construction of the test scenario
  2. Learn ROS
  3. Task allocation algorithm designation
  4. Implementation and validation
  5. Documentation

Bonus work packages:

  1. Joint research with UST-01B
  2. Paper writing for journals or top conferences

Please contact Ming Liu for details.

UST-06 Pointcloud-based semantic mapping using tensor voting frame (MA/SA)

UST06
UST06

Pointcloud is the basic representation for the state-of-the-art mobile robotic research. Comparing with computer vision, which has adequent information e.g. color and illumination for each pixel, pointcloud is only embedded with plain coordinate information. Tensor voting is a framework which helps to add additional semantic information beyong that. These additional semantic information will help full representation of an environment. This is the main target of the project.

Preliminary requirements:

  1. Linear algebra
  2. Programming: prefer C/C++, matlab

Core work packages:

  1. Learn Tensor Voting Framework
  2. Designation of semantic representation
  3. Implementation
  4. Documentation

Bonus work packages:

  1. Demonstration with mobile robots
  2. Paper writing for journals or top conferences

Please contact Ming Liu for details.

UST-07 Pointcloud-based scene classification and topological mapping (MA/SA)

Localization and scene classification are two basic fundamental functions for service robots. In this project, topological mapping and localization will be extensively studied. Specially, pointcloud will be used as the major represenation. Several clustering algorithms and learning algorithms will be conducted.

Preliminary requirements:

  1. Basic knowledge on machine learning and geometry
  2. Programming: C/C++

Core work packages:

  1. Learn ROS
  2. Designation and implementation for real robotic system
  3. Validation
  4. Documentation

Bonus work packages:

  1. Paper writing for journals or top conferences

Please contact Ming Liu for details.

UST-08 Riemannian navigation for mobile robots (BA/MA)

Various configuration spaces can be adopted to represent your data. For the navigation in semi-structured 2.5D environment, efficient representation of the surroundings is the key to calculate a target path in near real-time. When a pointcloud-based representation is used, we could consider each local region as a local plane, which is embedded with a local Riemannian tensor. Starting from this point, the navigation problem for mobile robots can be considered as a geodesic estimation problem on a Riemannian manifold. You are supposed to describe this structure and implement these concepts to real robotic system.

Preliminary requirements:

  1. Basic knowledge on geometry
  2. Programming: matlab, python or C/C++

Core work packages:

  1. Designation of Riemannian metric for local regions of pointcloud
  2. Optimization of the geodesic search algorithm
  3. Implementation and validation on datasets
  4. Documentation

Bonus work packages:

  1. Real-time implementation on real system
  2. Paper writing for journals or top conferences

Please contact Ming Liu for details.

UST-09 Large-scale localization using hand-held devices (BA/SA/MA)

For visitors, it is easy to get lost in public places such as railway stations. This project aims at developing an applicable system to localize in such large indoor environments via smart-phones (Android, iOS etc). 

The project is to be built from scratch, which means the student needs to first gather real data in target environments with ground-truth. Background knowledge of computer vision or experience of Android/iOS APIs is welcome to ensure the success of the project.

Preliminary requirements:

  1. Programming: C/C++, prefer Android/iOS APIs
  2. Computer vision and geometry
  3. Basic knowledge on machine learning

Core work packages:

  1. Acquisition of data-sets with ground-truth
  2. Android Interfacing for database access and visualization
  3. Efficient image retrieval, e.g. by bag-of-words or hash-tree
  4. Salient feature selection, e.g. by maximizing information entropy
  5. Localization by estimation of the transformation from the real world to image frame
  6. Validation
  7. Documentation

Bonus work packages:

  1. Visual guidance for visitors - real use-case
  2. Paper writing for journals or top conferences

Please contact Ming Liu for details.

UST-10 2D Localization using one-bit sensor (MA/SA/BA)

UST10
UST10

The goal of this project is to implement a low-bandwidth localisation algorithm on the inexpensive Thymio II mobile robot. The localisation algorithm itself will run on a PC, connected to the robot with a low-power wireless connection. The approach is to localise against a know black&white pattern using the two ground infra-red distance sensors, located on the front of the robot.

The localisation algorithm should employ a probabilistic filter, considering the know pattern, the command to the robot, and the observation. The filter should initially be implemented in numpy using dense probability tables, and, if time permits, then be implemented as a sparse particle filter. The work will use the open-source ROS robotics software framework.

Preliminary requirements:

  1. Programing: matlab, C/C++
  2. Basic knowledge on Kalman Filter and Particle Filter

Core work packages:

  1. Problem definition, robot and ROS discovery, and literature review
  2. Environment Setup and Dataset Gathering
  3. Characterisation of motion model
  4. Dense localisation coding
  5. Dense localisation validation

Bonus work packages:

  1. SLAM system
  2. Localization in arbitray binary pattern
  3. Paper writing for journals or top conferences

Please contact Ming Liu for details.

Comments