Machine-learning algorithm for robotic pressure-sensitive skin

Project Code: 10464909

Faculty: Auckland Bioengineering Institute

Department: Auckland Bioengineering Institute

Main Supervisor: Dr Samuel Rosset (sros143)

Co supervisors: Associate Professor Iain Anderson

Application open date: 29 Jun 2020

Application deadline: 30 Nov 2020

Enrolment information: NZ Citizens, NZ Permanent Residents


Safe and damage or injury-free interaction between conventional hard robots and delicate objects remains a challenge. This limits opportunities in elderly care, healthcare, or mechanised fruit harvesting, for which it is key to avoid injury, pain, or bruising. We propose to develop a soft compressible sensing skin that will provide robotic manipulators with touch sensitivity while also providing a compliant and deformable interface with the object they are manipulating. We will develop an ultra-soft capacitive touch sensitive sensor with a monolithic structure and a single pair of electrodes that, despite its simple structure, can inform on both the touch location and its force magnitude. The key to this technological advance resides in the high electrical resistance of the capacitive sensors stretchable electrodes, making it behave as an RC transmission line where signal attenuation is frequency dependent.  By interrogating the sensors response to a broad-bandwidth input voltage signal we are able to identify where the sensor is pressed, and the amplitude of the applied force. But this needs some further development before it can be useful.

This project has two objectives: 1) Sensor development, and 2) Construction of a test platform for the sensor where it will be exposed to large-bandwidth excitation, and development of a sensing algorithm to identify the amplitude and location of deformation.

Objective 1 is already underway. As a Master student, you will drive the second objective. Specifically, your task will be to enable the measurement of both pressure intensity and location on a long and thin compliant sensor, while using a single pair of electrodes. This will drastically reduce the complexity of the sensing surface; only requiring two wires. Unlike expensive soft sensing skins relying on a matrix of sensors, costly flexible transistors, or complex structures, you will use an inexpensive monolithic sensor, but rely on computational power to extract location information.

What we are looking for in a successful applicant

The project is multidisciplinary but would be most suitable to students with an electrical engineering, mechatronics, bioengineering, or engineering science background. The project will involve:

  • Lab-based fabrication of equipment that will include some  soldering and 3D printing.
  • Signal generation and acquisition: you will send an excitation signal with a large frequency bandwidth, and measure the current going through the sensor using computer-controlled equipment
  • Signal processing: you will process the acquired signals to remove noise and study the frequency spectrum
  • Modelling: you will use electrical circuit modelling to predict the effect of local deformation of the sensor on the electrical signal
  • Algorithm development: you will develop an algorithm to extract the amplitude of the deformation and the location where force is applied from the measured signals. One of your tasks will be to identify the best method (Machine learning, analytical model, etc.)
  • Programming: You will write code to control the signal sent to the sensor, acquire the sensor data, and run the amplitude and location identification algorithms
  • CAD and prototyping: You will need to design and prototype a few elements to assemble a testing bench and a demonstrator

 The objectives of your contribution to the project are centred on the development of 1D sensor with touch location. Specifically you will:

  • Assemble a testing bench comprising signal generation and data acquisition
  • Develop a method based on sending an electrical signal with a large frequency spectrum through a soft sensor compression sensor in order to identify where the sensor has been pressed, and how hard it has been pressed.
  • Apply your method to a 100 mm long 1D compression sensor, with the aim of detecting force amplitudes in the 0-5 N range, with identification of the zone of contact with a resolution of 10 mm.
Other information

Co supervisors: Prof. Iain Anderson and A/P Suranga Nanayakkara

This master project is part of a SfTI seed project. Fees and stipend are covered by the project. You will be working with one of our PhD students who will be developing the soft compression sensors, while you will develop the sensing algorithm.

  • Start of the project: Semester 1 2021 (March 2021).
  • International students: Unfortunately, the project only covers fees for a local student.
Related opportunities