Introducing ML/RL Success Stories
In this new blog series we’ll regularly take a look at publications that show the success of applying machine learning, reinforcement learning and robotics to real world commercial applications. These applications can be ones we find interesting, relevant to the field where DeepSim can be applied or just simply cool. In this first post in the series, we take a look at robotic manicures and a more efficient way of harvesting strawberries. Expect many more of these success stories as reinforcement learning gets applied more and more by a wide range of companies working on many different innovative products.
Manicure by robots
Manicures are a time consuming and precise process, and often a social activity, but StartUp Nimble is working to change this. They have developed new technology to have your nails painted, at home, using their automatic technology in just 10 minutes. Ideal for those last minute parties. The manicure device uses machine learning to accurately detect the nails. This allows the robotic brush to apply the paint at exactly the right location and with the precise thickness required to get a solid and strong layer. This is the type of automation I can totally get behind, although you lose the fun of hanging out with your friends at the salon.
Picking strawberries is no longer backbreaking work
Strawberries, who doesn’t love them right? Traditionally picking strawberries is a backbreaking task where the pickers are bent over or sitting down on the ground for hours on end while they make their way along the rows of strawberry plants selecting the ripe strawberries. This hard work results in labor shortages, and that in turn leads to unpicked fields and rotting strawberries. Now Traptic is trying to change all that. They have developed a fully automated robotic strawberry picker. The robot uses machine vision, powered by neural networks, to detect the ripe strawberries and then uses fully automatic robotic arms to pick the ripe strawberries. The arms need to be gentle enough to not squeeze the juicy ripe strawberries and nimble enough to curve around and underneath the leaves and unripe strawberries. A lot of training is required to teach AI to operate robotic arms, and platforms like DeepSim can be used for this training. The platforms train in a virtual world to prevent squeezing and destroying tons of strawberries in the beginning of the training phase. After the initial training the trained neural network agent can be fine tuned on the real hardware. Traptic now shows the results of such training and automation with the commercial deployment of their farming robot.
That’s all for this edition, we hope you enjoyed it. Please join us again next time as we continue to showcase new innovations!