Michael Clark: "Can a robot pick soft fruit: requirements and simulation"
Details
*** This will be an in-person, networking, and social event. Please RSVP for catering purposes. ***
Many soft fruit are still picked by humans. Can a robot do it instead?
In the first half of this talk lets look at the technical requirements, heavily drawing from this 2014 paper (https://drive.google.com/file/d/14RxXPjqyu5rSF2grh2mHVY9avmV9i5sx/view?usp=sharing) and 2010 paper (https://core.ac.uk/reader/38943040).
Once we have defined the task requirements, lets ask: Can reinforcement learning help with some of the challenging aspects? If it can't do it in a simulator, then it can't do it in the (more expensive) real world. Since no simulator exists, lets build one.
The second part of this talk will share the experience of building a simple pybullet simulator for apple picking. The simulator is basic but includes a random tree, and a robot that must navigate around the branches using a camera.
The code for the simulator will be released on github, in an unpolished form, for anyone who wants and is brave enough to use it.
Finally, we will brainstorm about why it didn't work. This part will likely have some good suggestions or at least some good heckling.
Speaker bio:
Mike has a MSc from Victoria University and a BSc (Hons) in Physics from the University of Canterbury in New Zealand. Before working in machine learning Michael Clark worked in oil and gas. He now works as a director at Three Springs Technology who focus on world class machine learning for medical imaging.
Rundown:
5.30 pm networking
6.00 pm Welcome and presentation
6.45 pm continue networking (snacks and beverage will be served)
8.00 pm event finishes, we hope to see you in the future event
Thanks to Three Springs Technology for sponsoring food and drinks for this non alcohol event.