Robots, new machines in our daily lives, diversify. Recent progress has made the rising of cobots possible. Cobots are robots which collaborate with human beings. Contrary to traditional robots, this new type of robot requires the expertise of an operator to run. The Learning from Demonstration creates an original way of programming. The operator can manipulate the robot’s arm in order to teach it the movement to realize. The present thesis proposes an improvement of this learning through these three axes: the data processing, the learning, and the acceptability. Before being used by the learning, data is retrieved during the kinesthetic demonstration, then temporally aligned, and filtered to improve its quality. A novel learning algorithm ...