School holidays, machine learning and robots that are learning to “see”
This past school holiday break, Coding Kids took a group of school children to visit the Australian Centre for Robotic Vision at QUT, Gardens Point Campus in Brisbane. The Centre specialises in research and development of robots learning to ‘see’. Their ability to see is the remaining technological roadblock to ubiquitous deployment of robots into society…
The Australian Research Council Centre of Excellence for Robotic Vision is playing a key role in overcoming this roadblock. The Centre is developing the underlying science and technologies that will enable robots to see to understand their environment and to perform useful tasks in the complex, unstructured and dynamically changing environments in which we live and work.
In our tour of the research facility we learned about how it is difficult for robots to “see” and recognise images the way that humans do and that there is a lot of research being undertaken in this area so that improvements can be made to how we use technology. The potential for robots to help us and our communities is endless if robots could learn to see like humans. Robots, and machines, only understand 1’s and 0’s, so how do roboticists and scientists teach and train robots to see images?
One of the robot prototypes that is being developed is Guiabot, an autonomous driving vehicle that is learning to “see” its environment. Unlike humans, robots find it more difficult to determine whether two slightly different images are of the same location. The students played a game against a robot to test how good they were at identifying whether two images were of the same location. One student had a go at the game and obtained a score of 10/10. The robot’s score of the same game was only 9/10. This showed that it is much easier for humans to identify whether two photos are of the same location. Although the technology is improving, robots are still not able to recognise their location visually as well as humans.
We learned about robots that were learning to “see” using a process called machine learning. In 1959, Arthur Samuel defined machine learning as a “Field of study that gives computers the ability to learn without being explicitly programmed”.
Here are the three robots at Robotic Vision that are learning to “see” using machine learning:
- COTSbot is a robot that is learning to “see” and identify crown-of-thorns starfish (COTS) in the Great Barrier Reef. The COTSbot is designed to control the overpopulation of COTS which is damaging the Great Barrier Reef.
- Harvey the harvester by QUT (based on a UR5 arm) is a capsicum picking robot. It is learning to “see” and identify ripe capsicum that is ready for picking. 30% of crops in Australia are wasted due to lack of workers at the exact time of year when crops need to be harvested.
- AgBot by QUT is a 3m wide weed remover robot that can identify the cotton plant and weed plants which require removal.
These three robots are learning to identify a specific object so that it can do useful work for us. The method of learning and training the robots is called machine learning. Machine learning is based on a complex set of algorithms which computers, or in this case, robots, use to learn from and make predictions using known data.
The robots first start their learning journey by learning to identify their target objects in images, e.g. COTS, capsicums and weeds. The robots then progress to learning to identify 3D printed versions of the target objects. After succeeding at this level of training the robots then start their training to identify the real life objects e.g. COTS in the Great Barrier Reef, ripe capsicums in orchards and weeds in the soil. At this level of training the robots learn to differentiate between 3D print versions and real target objects.
The Australian Centre of Excellence for Robotic Vision specialises its research and development for robots to ‘see’ and plays a key role in overcoming the last remaining technological roadblock for ubiquitous deployment of robots into society. The potential for robots to improve our lives and the environment is limitless. This is an interesting field of research and our school group visit to the centre was a good opportunity to engage children with the wonders of technology. Just as children are learning, so to are the robots.
Here are recommended YouTube videos to learn more about machine learning, neural networks and deep learning.
- Learn more about machine learning.
- Learn more about how neural networks, a machine learning technique, is used to train machines to “see” images.
- Learn more about deep learning, a branch of machine learning.
Like our Facebook page to keep up to date with our latest free tech event for kids.