http://www.digitaljournal.com/tech-and-science/technology/how-to-train-a-robot-to-feed-you-dinner/article/545985

How to train a robot to feed you dinner

Posted Mar 23, 2019 by Tim Sandle
A new robot has been developed to assist people with eating. The robot can pass food to people who are ill, injured or elderly, or otherwise require support. The robot has an algorithm that guides to selecting the required food item.
A succulent cooked steak  prepared at The Strand Palace Hotel Carvery.
A succulent cooked steak, prepared at The Strand Palace Hotel Carvery.
The meal-assisting robot has been developed by researchers at the University of Washington. The robot uses a specially developed object-detection algorithm termed RetinaNet. The algorithm enables the robot to scans a plate of food and then to identity the types of food on it and places a frame around each item. The robot will pass the require food item to the person using the robotic meal-assistant.
A large number of adults require assistance with eating. For the U.S. this could be up to one million people, according to Futurity. For those supporting such people, this can be a tiresome task and one that is open to automation.
Discussing this, lead researcher Dr. Siddhartha Srinivasa states: "Being dependent on a caregiver to feed every bite every day takes away a person’s sense of independence. Our goal with this project is to give people a bit more control over their lives."
To develop the robot, the researchers used machine learning. While robotics has advanced sufficiency for a machine to pick up and pass food, getting the robot to select the required food was something more challenging. For the training, the scientists arranged plates with a dozen different kinds of food, which ranged in consistency from hard carrots to soft bananas. The plates also included foods like tomatoes and grapes.
Using human volunteers first, the researchers gave people a fork and asked them to select different pieces of food , to pick them up and to pretend to to feed the food items to a mannequin. The fork was fitted with a sensor and the data from this was used to train the robot. In time the robot used the same force-and-skewering strategy to pick up all the pieces of food.
The video below shows the robot in action:
The research was presented at the ACM/IEEE International Conference on Human-Robot Interaction in South Korea, which took place during March 2019.
The development of the robot is described in the journal IEEE Robotics and Automation Letters. The research paper is titled "Towards Robotic Feeding: Role of Haptics in Fork-Based Food Manipulation."