Deep Learning for Robotic Grasping
Lead Research Organisation:
University of Surrey
Department Name: Vision Speech and Signal Proc CVSSP
Abstract
This project looks at using deep learning and computer vision techniques to aid robotic grasping in warehouse situations. This includes using deep reinforcement learning methods to learn how to grasp unknown objects and using vision based deep learning to identify instances of unknown objects for grasping and manipulation. The novel research will be around the learning methods and how to use them with robotics given the low quantity of available data and the complications of integrating it with the hardware.
Organisations
People |
ORCID iD |
| Richard Bowden (Primary Supervisor) | |
| Rebecca Allday (Student) |
Studentship Projects
| Project Reference | Relationship | Related To | Start | End | Student Name |
|---|---|---|---|---|---|
| EP/N509772/1 | 01/10/2016 | 30/09/2021 | |||
| 1817199 | Studentship | EP/N509772/1 | 01/10/2016 | 31/03/2020 | Rebecca Allday |
| Description | The work funded by this grant has so far focused on the use of reinforcement learning for robotic manipulation. There has been a lot of focus around the separation of control and perception tasks - in order to get the most out of the deep reinforcement learning and deep learning techniques for vision. |
| Exploitation Route | This work can be taken forward in industry to be applied to warehouse picking and to other robotic manipulation tasks. It could also be used in academia to continue to improve perception and control problems. |
| Sectors | Agriculture, Food and Drink,Digital/Communication/Information Technologies (including Software),Electronics,Retail |
| Description | Tesco Labs Sponsorship |
| Amount | £30,000 (GBP) |
| Organisation | Tesco Plc |
| Department | Tesco Labs |
| Sector | Private |
| Country | United Kingdom |
| Start | 10/2016 |
| End | 09/2019 |
| Title | Research Code |
| Description | An extensive body of research code for perception and reinforcement learning for robotics. |
| Type Of Technology | New/Improved Technique/Technology |
| Year Produced | 2019 |
| Impact | Still on-going work - difficult to describe impact. |
| Description | Applicant Days |
| Form Of Engagement Activity | Participation in an open day or visit at my research institution |
| Part Of Official Scheme? | No |
| Geographic Reach | National |
| Primary Audience | Other audiences |
| Results and Impact | Showing demos and talking about the work we do in this project as part of the applicant day for prospective undergraduate students. They are mainly 17/18 year old from schools around the country who are considering studying electronic engineering. They also often bring their parents with them on the tours. There are always lots of questions about robotics and how to get involved with machine learning. The students are always excited by the prospect of getting the opportunity to work in the robot lab as part of their undergrad and see research as a potential career option. The feedback from the academic in charge of applicant days is that the tour of the robotics lab is always a highlight of the day with many students enthusiastic about studying electronics and robotics. |
| Year(s) Of Engagement Activity | Pre-2006,2017,2018,2019 |
| Description | Talk at BMVA Technical Meeting |
| Form Of Engagement Activity | A talk or presentation |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Postgraduate students |
| Results and Impact | Talked at the BMVA Technical Meeting on "Robotics meets Semantics: Enabling Human-Level Understanding in Robots". The talk was titled "Reinforcement Learning For Human-Inspired Grasping". This was great for reaching industrial professionals who attended the event and other PhD students. |
| Year(s) Of Engagement Activity | 2018 |
| URL | https://www.eventbrite.co.uk/e/bmva-technical-meeting-robotics-meets-semantics-enabling-human-level-... |
| Description | Tesco datamashup |
| Form Of Engagement Activity | A talk or presentation |
| Part Of Official Scheme? | No |
| Geographic Reach | National |
| Primary Audience | Industry/Business |
| Results and Impact | Talked about "Deep Learning Based Robotic Grasping For Warehouse Picking" at a breakout session at Tesco's internal data science conference (datamashup). Discussed the work we have done with a lot of data scientists at Tesco from many different areas including Tesco Bank, Tesco Labs and Clubcard. |
| Year(s) Of Engagement Activity | 2018 |