2024 MSc Human and Biological Robotics

& 2024 MSc Communications and Signal Processing

​

Tactile Perception and Robotic Hand Design:

​

Students:

Louis Flinn; Syed Hasan; Julian Ing

Project Description: 
In-hand manipulation involves adjusting the position and orientation of an object within the palm and fingers. This MSc project aims to design and prototype an affordable robotic hand that not only mimics the complex motions required for in-hand manipulation but also integrates tactile feedback to enhance manipulation capabilities. This project requires the student to push the boundaries of current robotic hand designs by incorporating tactile sensors and developing algorithms to interpret and respond to tactile data, thus enabling the hand to perform delicate and skillful in-hand manipulations.

​

Project Goals:
i)    To design a dexterous robotic hand with articulations that reflect the degrees of freedom necessary for in-hand manipulation; employ cost-effective manufacturing processes and materials, potentially including 3D printing, to construct the hand.
ii)    To integrate affordable tactile sensors that provide critical feedback for in-hand manipulation, such as slip detection and pressure distribution; map sensor placements effectively across the hand, focusing on areas that maximize manipulation control.
iii)    To leverage open-source electronics and software platforms to manage costs and encourage community collaboration and innovation; enable the construction of a compact actuation and control system that can process sensory input and coordinate the actuation of the hand with precision and responsiveness.
iv)    To construct a working prototype that embodies the designed features and can be tested for functional capabilities; conduct a comprehensive series of tests that assess the in-hand manipulation skills of the robotic hand, including object reorientation, precision lifting, and controlled release.

​

Required Skills:
i)    Proficient knowledge of mechanical engineering principles and experience with 3D CAD modeling.
ii)    Experience with sensor technology and skills in electronic circuit design and familiarity with microcontroller programming and interfacing.
iii)    Practical experience in prototype building, including familiarity with rapid prototyping techniques.

​

Expected Outcomes: 
i)    A functioning prototype of an affordable robotic hand capable of in-hand manipulation with tactile feedback.
ii)    A set of test results demonstrating the hand's manipulation capabilities and the effectiveness of the tactile feedback system.
iii)    A final project report detailing the design process, control algorithms, technical challenges, performance evaluation, and recommendations for future development.

​

​

​

Sim-to-Real Transfer Learning for Robotic Manipulation:​

​

Students:

Yudong Mao; Zhenyang Tao

​

Project Description: 
The objective of this MSc project is to leverage the potential of simulation environments for training deep learning models for controlling robotic arms for dexterous manipulation. However, the deep learning models that drive these robotic manipulators necessitate vast amounts of experimental data for training, which can be impractical and economically infeasible. To this end, this project will focus on developing a robust sim-to-real transfer learning framework that allows a pre-trained model in a simulated environment to adapt effectively to a real-world setting.
The student will train a deep learning model to accurately identify and manipulate target objects in a virtual domain. This training will be performed using state-of-the-art simulation software that provides a high-fidelity approximation of real-world physics and environmental conditions. The core challenge lies in bridging the gap between the simulated and real environments. This project will not only address the practical challenges of training robotic manipulators but will also contribute to the broader understanding of transfer learning techniques, which are crucial for the deployment of intelligent systems in dynamic real-world scenarios.

​

Project Goals:
i)    To develop a deep learning model capable of object detection and grasp point selection within a simulated environment.
ii)    To conduct model training within the simulation platform, utilizing synthetic data generated within the simulated environment and using advanced deep learning techniques that are resilient to the domain shift between simulation and reality.
iii)    To implement and fine-tune sim-to-real transfer learning algorithms to adapt the simulation-trained model to real-world conditions effectively.
iv)    To evaluate the performance of the transferred model on a physical robotic manipulator setup and iteratively refine the model through real-world training cycles to enhance accuracy and reliability.

​

Required Skills:
i)    Proficiency in deep learning frameworks such as PyTorch, TensorFlow, or Keras, with the ability to construct and train deep neural network architectures.
ii)    Knowledge of computer vision techniques and libraries, particularly OpenCV, for processing and analyzing visual data from both simulated and real-world environments.
iii)    Experience with robot simulation software, capable of simulating physics, robot kinematics, and environmental interactions to generate synthetic training data.

​

Expected Outcomes: 
i)    A deep learning model trained in a simulated environment for robot manipulation.
ii)    A tested and validated transfer learning pipeline for adapting simulation-trained models to the physical robot.
iii)    A comprehensive evaluation report detailing the model's performance, with insights into the challenges of sim-to-real transfer and potential solutions.