Developmental algorithms for robotics: Understanding the world of objects, interactions and tools.

Lead Research Organisation: Aberystwyth University
Department Name: Computer Science

Abstract

This research project is a psychologically-inspired investigation of an analogy of infant play as the central mechanism for autonomous, self-motivated robots that learn the local physics of their world.

We note that infants and children at play exhibit exactly the kind of autonomous learning that would be very desirable in robotics. Infant play has a major role in the acquisition of new skills and cognitive growth. Noticing that early infants spend hours in play, we have designed a computer analogy of infant play and this project is an in-depth investigation into the use of play as a means of building subjective understanding of the physics of the local world.

The project will implement a play generator algorithm on an iCub humanoid robot and perform experiments with a wide range of scenarios involving varieties of objects. This includes playing solitarily with objects to learn their properties, and interactive play with a human participant. We also include experiments with tool use (using one object as a tool for acting on another) to investigate how objects may become extensions of self.

A panel of selected scientific experts on infants and play will provide their psychological expertise throughout the project and will also assist with the design of a series of matching experiments that will compare results from the robot model with those from selected psychological experiments on infants.

The data from the experiments will be analysed and interpreted to shed light on a set of scientific issues. When we report on the results we will also extract some general principles for robot learning through play. We will examine the applicability of these principles in new robotic and intelligent systems developments. For example, we anticipate particular applications in areas such as assistive technology and home care where the re-programming of mass-produced systems is not feasible. We believe technology with a developmental approach will have wide implications and provide an alternative to "building robots" by establishing the idea of "developing robots" for applications.

Planned Impact

The scientific output of this project will produce new algorithms and methods for a form of robot learning based on development. The knowledge and understanding gained will be interpreted in terms of principles for autonomous operation and the safe handling of novel experiences. Apart from the academic beneficiaries, we see scope for considerable societal and economic impacts.

There are three immediate fields that will benefit: (1) the robotics and associated manufacturing industries; (2) the healthcare, medical, and domestic care sectors; and (3) science education and public awareness.

(1) Industrial robotics is now a mature field but most robot devices in real-world environments are still controlled by either fixed programs or human operators. Examples include robot assistants, search and rescue robots, and surveillance and military robots. Unlike factory robots, it is often the case that real-world robots cannot be programmed to cover every task contingency. Each new task scenario demands knowledge appropriate to that specific case and it would be desirable for the robot to accumulate this knowledge rather than rely on the operator every time. Our results will address how this can be achieved and how operators may shape or train robots to improve their performance, thus achieving some operational autonomy, at least for part of the task. Such autonomy is very desirable for reducing operator stress and costs, increasing functionality, and extending application areas. This could provide a technological advance that has major benefits for robot product innovators, commercial robotics companies, and application developers.

(2) Assistive technology is a current world priority and a major challenge, driven by the demographic shift towards care for the aging population. This crucial need for flexible, adaptive technology involves all kinds of assistive, service, healthcare and domestic applications. As for (1) these are all real-world robotics scenarios that require flexibility in coping with the physical world, and the capability to adapt to the many and varied domestic needs and personal environments. Providing some autonomy in such domestic support devices should have a considerable impact on the quality of people's lives and bring real benefits to a range of organisations both public (e.g. NHS) and private sector (service and technology providers). The benefits will include; releasing demand for care staff, increasing the independence of users, generating new products and service industry possibilities, and increasing UK competitiveness in the service and healthcare domains.

Our project partners, QinetiQ, will be actively engaged in identifying new application opportunities in both the above fields. They are especially interested in new ways of task configuring, calibrating and controlling robotic systems that are cheaper to implement and more robust.

(3) Our use of the local Technocamps organisation (http://www.technocamps.com/) will require new materials to be prepared for school pupils aged 11-19. The Technocamps staff will prepare these and this will result in the availability of several workpackages of material. These will have wide applicability for public awareness of science promotions and will be made openly available.
 
Description Developed a model of how infants visual perception changes over the first year of life and how this helps to shape their learning, both of the world around them and their fine motor skills. Using this result, we can measure the impact the model has on an infant's ability to recognise and distinguish different objects at different stages in their development. Early on the limited vision means all objects look pretty similar, with just the bright or moving features being distinguishable. As vision gradually develops it is possible to distinguish an increasing number of different colours and eventually the edges of objects. Combining in information from the feel and hand shape for holding objects gives a rich description for distinguishing multiple objects.

Throughout their development, playful behaviour acts as a key driver for learning, leading to the discovery of new skills and an understanding of their environment. Modelling this on a robotic platform shows the benefits of play, along with the ability to adapt and transfer learning from one scenario to another. All of this is linked into psychological studies of infant behaviour with models that may help to understand the internal mechanisms and developmental changes resulting in behaviour changes in infants.
Exploitation Route Continuing to discuss possibilities with a leading AI company from America who are keen to apply our models for learning skills and coordination as part of the robotic system they are developing.
Sectors Aerospace, Defence and Marine,Digital/Communication/Information Technologies (including Software),Healthcare

URL https://www.aber.ac.uk/en/cs/research/ir/robots/icub/dev-icub/
 
Description See public engagement activities.
 
Description Aberystwyth University PhD Scholarship
Amount £45,000 (GBP)
Organisation Aberystwyth University 
Sector Academic/University
Country United Kingdom
Start 03/2015 
End 02/2018
 
Title Developmental timeline 
Description Detailed developmental timeline covering development of perception and growth of object understanding over first two years of an infants life. 
Type Of Material Improvements to research infrastructure 
Year Produced 2017 
Provided To Others? Yes  
Impact See publications 
 
Title Early recognition of multi-modal objects 
Description Technique for enabling early recognition of multi-modal objects, bringing together psychological insights on protoobjects. This also enables an early form of object categorisation based on similarity in form, with the ability to categorise by function learnt through playful interaction. 
Type Of Material Improvements to research infrastructure 
Year Produced 2017 
Provided To Others? Yes  
Impact See relevant publications. 
 
Title Grasp and tactile representation 
Description Grasp and tactile representational techniques that can be used to distinguish between objects by feel alone 
Type Of Material Improvements to research infrastructure 
Year Produced 2017 
Provided To Others? Yes  
Impact See relevant publications. 
 
Title Play as a driver 
Description Play as a driver for robots learning new skills and gaining new knowledge about the world. 
Type Of Material Improvements to research infrastructure 
Year Produced 2016 
Provided To Others? Yes  
Impact Play as a driver for robots learning new skills and gaining new knowledge about the world. 
 
Title Data and models 
Description No legal or ethical constraints affecting the models or data. Models are described in the external publications. Datasets from models are available at: http://icub.dcs.aber.ac.uk/index.php/publications or https://www.aber.ac.uk/en/cs/research/ir/robots/icub/dev-icub/ 
Type Of Material Computer model/algorithm 
Year Produced 2016 
Provided To Others? Yes  
Impact Observed downloads and citations. 
 
Description QinetiQ 
Organisation Qinetiq
Country United Kingdom 
Sector Private 
PI Contribution Robotic systems.
Collaborator Contribution Scientific advisory committee member.
Impact N/A.
Start Year 2015
 
Title Developmental approach software 
Description Software facilitating a developmental approach to learning, with experimental results investigating the developmental time line in more depth than has previously been considered. 
Type Of Technology Software 
Year Produced 2018 
Impact See publications 
 
Title Mapping software 
Description Significant extension to previously developed mapping software, modelled on layers of neurons in the brain, that enables greater detail and breadth of representation and connection for modelling learning in infants. 
Type Of Technology Software 
Year Produced 2017 
Impact See publications 
 
Title Play generator 
Description The play generator mechanism has been significantly developed to enhance the scope for the learning and generating of its own goals for directing play. 
Type Of Technology Software 
Year Produced 2016 
Impact See relevant publications. 
 
Title Play mechanism 
Description The play mechanism is opening up new avenues for exploration of open-ended learning, including a hierarchical model of skill development for developing more complex actions and understanding. 
Type Of Technology New/Improved Technique/Technology 
Year Produced 2017 
Impact See publications. The detailed model enables a wide range of experiments to compare against psychological studies, to help understand the developmental changes observed in infants. 
 
Title Proto-object representation 
Description Proto-object representation for learning multi-modal object features. 
Type Of Technology New/Improved Technique/Technology 
Year Produced 2016 
Impact See publications 
 
Description Aberystwyth Beach Lab 2017 
Form Of Engagement Activity Participation in an activity, workshop or similar
Part Of Official Scheme? No
Geographic Reach Regional
Primary Audience Public/other audiences
Results and Impact Research promoted to general public during Aberystwyth Beach Lab 2017, receiving over 1500 visitors of all ages.
Year(s) Of Engagement Activity 2017
 
Description Aberystwyth Robotics Week 2018 
Form Of Engagement Activity Participation in an activity, workshop or similar
Part Of Official Scheme? No
Geographic Reach National
Primary Audience Public/other audiences
Results and Impact Over 3000 visitors during the course of the week.
Year(s) Of Engagement Activity 2018
 
Description EPSRC Pioneers, Advances Wales, BBC 
Form Of Engagement Activity A magazine, newsletter or online publication
Part Of Official Scheme? No
Geographic Reach International
Primary Audience Public/other audiences
Results and Impact Research featured in EPSRC Pioneers and Advances Wales magazines, as well as on BBC.
Year(s) Of Engagement Activity 2016
URL http://www.bbc.co.uk/programmes/b08fwtlt;https://www.epsrc.ac.uk/newsevents/pubs/pioneer17/;https://...
 
Description EPSRC photo competition 2014 
Form Of Engagement Activity A press release, press conference or response to a media enquiry/interview
Part Of Official Scheme? No
Geographic Reach International
Primary Audience Public/other audiences
Results and Impact EPSRC photo competition 2014 - 1st place in People category, promoting outstanding UK engineering and physical sciences.
Year(s) Of Engagement Activity 2014
URL https://www.flickr.com/photos/epsrc/15752505647/in/album-72157651335202305/
 
Description EPSRC photo competition 2015 
Form Of Engagement Activity A press release, press conference or response to a media enquiry/interview
Part Of Official Scheme? No
Geographic Reach International
Primary Audience Public/other audiences
Results and Impact EPSRC photo competition 2015 - 1st place in People category, second year in a row, promoting outstanding UK engineering and physical sciences.
Year(s) Of Engagement Activity 2015
URL https://www.flickr.com/photos/epsrc/24309680141/in/album-72157666782307310/
 
Description EPSRC photo competition 2016 
Form Of Engagement Activity A press release, press conference or response to a media enquiry/interview
Part Of Official Scheme? No
Geographic Reach International
Primary Audience Public/other audiences
Results and Impact 3rd place in people category, following two years in first place.
Year(s) Of Engagement Activity 2017
URL https://www.flickr.com/photos/epsrc/32535387495/in/photolist-pZZH7X-Rz3nbP
 
Description TV show 
Form Of Engagement Activity A broadcast e.g. TV/radio/film/podcast (other than news/press)
Part Of Official Scheme? No
Geographic Reach International
Primary Audience Media (as a channel to the public)
Results and Impact ITV "Wales This Week: My Robot Helper" on 30th Apr 2018
Year(s) Of Engagement Activity 2018
URL https://www.itv.com/walesprogrammes/wales-this-week
 
Description Welsh Government PR campaign 
Form Of Engagement Activity Participation in an activity, workshop or similar
Part Of Official Scheme? No
Geographic Reach National
Primary Audience Policymakers/politicians
Results and Impact Research to feature in Welsh Government PR campaign
Year(s) Of Engagement Activity 2018