Laboratory Notes

Machine Learning Control of Underactuated Mechanical Systems

Russ Tedrake
Computer Science and Artificial Intelligence Laboratory

Robots today move far too conservatively, using control systems that rely on maintaining full control authority at all times. Humans and animals achieve far superior performance by routinely executing motions that involve a loss of instantaneous control authority. Solutions in the literature for this “underactuated” control problem are few and problem-specific.

Walking robots designed with special attention to the learning control system by CSAIL Prof. Russ Tedrake and his group

Walking robots designed with special attention to the learning control system by CSAIL Prof. Russ Tedrake and his group

Recent advances in machine learning for control have made it possible to consider a more general solution to the underactuated control problem based on approximation techniques. In the Robot Locomotion Group, we are extending these techniques to develop a breed of simple mechanical systems designed to achieve extraordinary dynamic performance. Pictured above a biped, with a single motor at the hip, uses a nearly optimal control system allowing it to walk over a surprising amount of terrain.

The robotic bird, pictured below, uses flexible flapping wings to trade off the efficiency of an airplane with the maneuverability of a helicopter. This bird, which is already flying around Killian court, will be able to catch (bright orange) bread-crumbs out of the air and land on a perch.

A key feature of our robots is that we spend a considerable amount of time on mechanical design. Our walking robots are capable of walking stably down a small ramp without any actuators, and our birds glide passively before they learn to fly. These mechanical design elements have a dramatic effect on the performance of the learning control system, and ultimately govern the dynamic capabilities of the robots.

A robotic bird designed by the Tedrake group in CSAIL to use flexible flapping wings to trade off the efficiency of an airplane with the maneuverability of a helicoptor.

A robotic bird designed by the Tedrake group in CSAIL to use flexible flapping wings to trade off the efficiency of an airplane with the maneuverability of a helicoptor.

Pages: 1 2 3 4 5 6

Leave a Reply

Replies which add to the content of this article in the EECS Newsletter are welcome. The Department reserves the right to moderate all comments. If you would like to provide any updated information please send an email to newsletter@eecs.mit.edu.