The work attempts to find ways of modelling human motion patterns mathematically and then getting a robot arm to obey those mathematic rules in a safe and efficient manner. A good chunk of these rules dealt with producing optimal motion (in terms of 'effort') that could also deal with the robot having a imperfect model of its own body (for example, we didn't model the friction in the gearboxes). All the control is done dynamically (in terms of forces and torques) rather than kinematically (in terms of positions and angles). This is more difficult but gives more control over the robot.
The below figure shows one of the early results in the thesis, where the robot's reaching motion (to a target height) is based largely on effort minimisation and ends up appearing very human. The person in the image is only included for comparison, the robot isn't copying him.
You can view or download the full pdf of the thesis here.
No comments:
Post a Comment