Researchers try to help robots learn to use soft materials such as ropes

Researchers from the University of Michigan work make robots more functional and capable in the real world. Researchers who worked on the study said robot models were used to carry out tasks usually work well in a structured environment in the laboratory. However, when robots are used outside the laboratory, some of the most sophisticated models have proven inadequate in certain situations. The situation that proved very difficult for robots to do work with soft materials such as ropes or fabrics.

Researchers at Michigan University have created a way for robots to predict when they cannot trust their operating models and to recover when they find a model that regulates their control. Basically, the team is looking to try and teach the robot to do what it has, according to Robotics Ph.D. Peter Mitrano students. The purpose of the researchers is to have a robot can take everything and move it without knowing physics or geometry everything.

They created a simple model of dynamics rope while moving around open space. The team adds obstacles and creates a classifier that is learned when a simple rope model is reliable but does not try to understand more complex behavior about how the rope interacts with objects. The team then added a recovery step if the robot experienced a situation, such as the rope collided with obstacles, and the classifier determines the simple model is not reliable.

The researchers said their approach attracted inspiration from the fields of science and other robotics, where a simple model was still useful. By using a simple model of the rope, the team develops a way to ensure an object is being used in a suitable situation where the model is reliable, allows the robot to generalize its knowledge in a new situation that has never been found.