Science and Technology

Feedbot – A symbiotic robotic arm for meal assistance

In Portugal, thousands of people need assistance with their meals. Robotics can help, as Feedbot shows. It is a robotic arm designed to empower people, with cerebral palsy or other paralysis caused by degenerative diseases, to feed themselves autonomously. After perceiving the environment, Feedbot tracks the head of the user, and cooperatively places the food in a comfortable and safe position for the user to eat up the meal. The project is led by Manuel Marques, researcher at ISR/IST. The principal co-researchers are professor Nuno Nunes from Interactive Technologies Institute (ITI), professor João Paulo Costeira, researcher at ISR, and professor Manuela Veloso from Carnegie Mellon University (CMU) who currently leads the JPMorgan’s AI research team.

The desire to develop Feedbot is not only a consequence of the will of helping others or the passion for computer vision combined with robotics, it is also the result of a need felt by Manuel Marques himself. “The idea came up during my stay at Carnegie Mellon University with professor Manuela Veloso”, recalls Manuel Marques. “The problem has always existed: I can’t have lunch or dinner autonomously and we realised that both Robotics and Computer Vision could play an important role and contribute to an innovative solution”, says the ISR researcher.

Thus, through the combination of trajectory generation algorithms, developed at CMU, and 3D perception algorithms, developed at ISR / Técnico, the robotic arm started to gain the ability to move and adapt to the needs of future users. “Instead of making predefined movements, the robot arm adapts to each user’s movements and identifies, for example, whether the user wants to eat or talk. This allows not only to feed the user, but also to transform the meal into a social act, that is, interact with other people ”, explains Manuel Marques.

“Feedbot is already capable of feeding a person autonomously and I have already done that several times”, says Manuel Marques, who also adds “the algorithms used in Feedbot are of enormous importance, as they allow the use of low-cost hardware and, at the same, to overcome lack of precision”.

The testing phase of this project proved to be crucial to meet the goal of improving knowledge of human movement based on image recognition and motion analysis. “We developed a methodology that allows us to classify different patterns of head and trunk movements from videos. For example, we are able to identify people with large amplitude movements, but with little precision or, on the contrary, little amplitude, but great precision. This methodology was validated for users of CRPCCG [Rehabilitation Centre for Cerebral Palsy Calouste Gulbenkian]”, stresses the project leader.

Feedbot is designed to people with cerebral palsy, Parkinson’s patients, and other neurological diseases affecting upper limb movements. “In addition, and because the robotic arm adapts to each person’s movements, it is also suitable for those who have suffered a stroke or different types of accidents and are unable to use their hands and arms temporarily,” says Manuel Marques. In short, Feedbot can be a friendly arm for those who have difficulties with upper limb movements, either permanently or temporarily.

At the moment, the team has two Feedbots operating: one at Técnico and another at CRPCCG, managed by Santa Casa da Misericórdia de Lisboa (SCML), which is collaborating directly with the project. Making the robotic arm accessible to the largest number of people is the primary team objective, “because it’s very important for their autonomy and independence”, says Manuel Marques.

The project has already won an honourable mention by Exame Informática – “The Best of Technological Portugal”, in the “Software” category. Manuel considers this honourable mention very importante because “it recognises the relevance of the problem and the proposed solution”.