Biped Robot MABEL Runs Free!

University of Michigan's MABEL runs free for over 110 steps! In our opinion, this is the most realistic, human-like running achieved on a robot. It has a very satisfying feel to it. The robot just moves right. It is up in the air for more than a third of the duration of step. The height off the ground is right. Whereas other robots had their feet maybe one sixth of an inch off the ground, MABEL is 3 to 4 in inches in the air. The motion of the hip, which is like a bouncing ball, and the pitching of the torso give you the sensation of running. It all just makes you say, that is running.

Feedback algorithm for running was designed by Koushil Sreenath as part of his PhD dissertation. The detailed model used in the work was developed by Hae-won Park as part of his PhD dissertation.

For the feedback control aficionados, we used a nonlinear, compliant hybrid zero dynamics controller with active force control, running in real-time. How about that! MABEL weighs over 65 Kg, has a heavy torso (40 Kg), has point feet, and a cable-driven transmission system with compliance. This makes it a challenging machine to control. The Hybrid Zero Dynamics (HZD) framework was instrumental in our success.

The achieved peak speed is 3.06 m/s (6.8 mph), with an average speed of 1.95 m/s (4.4 mph). The obtained gait has flight phase that's almost 40% of the gait, with a ground clearance of 3-4 inches.


How did we do it?

The answer is not as simple as "we got one thing right". Our success arose from a combination of things, several of which are very technical, but if we had to focus on two primary things, it would be very good machine design and very good feedback algorithm design. Specifically, the coordination of those two aspects. The machine design determines the passive behaviors of the robot, or how it will move when all power is turned off; springs, masses, and so on will enable or limit what you can do with the feedback control. Therefore, the machine was designed with the intent to emulate some aspects of human biomechanics; we then created an extremely detailed mathematical model of the robot after it was built, and then designed our control algorithms around this very precise model. This approach is rare in robotics, and has not been accomplished in past bipedal running robots.

Machine design: The bipedal robot MABEL was designed in 2006-2007, and built in 2008. The novelty was to have a machine with a roughly human weight distribution and springs that act like tendons in the human body.

Human weight distribution means that most of the weight of the robot is concentrated in the torso (upper body), while the legs are relatively light, so they can be moved forward and backward quickly for fast locomotion.

The springs in the robot serve two purposes. The first purpose is that when the robot's legs strike the ground, the springs act as shock absorbers. Specifically, running has a flight phase, where both feet are off the ground, and a stance phase, where one leg is on the ground. When a 145 pound (65 Kg) robot like MABEL ends the flight phase by landing on a leg, the force is pretty large. The springs make the landing more gentle. In some sense, this is what the arch in your foot does for you, or a good pair of running shoes. The second purpose of the springs is to store energy. This is analogous to a pogo stick, where the robot bounces up and down on the springs, storing and releasing energy with each stride. This effect has been shown to be an important aspect of all animal running. MABEL seems to be the first robot with human-like morphology to get this right.

Feedback Control:

What is feedback? Most everyone has an intuitive notion of feedback, such as when a supervisor provides feedback on an employee's performance, or when your body regulates your temperature to a constant 98.6 F (37 C) despite varying levels of physical activity and outside temperature. Feedback means that the input signals that are regulating a system are adjusted as a function of measurements (observations) of the system.

MABEL has four electric motors, two for each leg, which provide power. Whether the robot is walking, running, or just standing, there is a feedback controller on a computer that measures all of the positions of the robot's joints and the angle of its body, and then determines the proper power commands to send to the motors.

The foundation for our feedback controller is the detailed mathematical model of the mechanism: we have used this model to determine the best relationship between the measured leg angle relative to the ground, and the motions of all other robot joints. Our feedback controller implements this specific relationship on the robot, using information from sensors to control the motors. The resulting motions, in conjunction with the springs and masses of the robot mechanism, determine the forces that the leg applies to the ground, realizing a running gait.

• Category

  Science & Technology