Wait, Stewart Platforms Can Walk Now?!

A Stewart platform is a nifty kind of parallel robot that uses six linear actuators, with ball joints at each end, to move the top with six degrees of freedom. That’s really useful if, for example, you want a platform that you can actuate to always remain level. In most implementations, the base is fixed as a rigid body. But James Patten of JD Robotics reimagined the traditional Stewart platform, ditching the fixed base and using the linkages as legs for ambulation on the Stewy robot.

To understand how this works, first picture a standard Stewart platform. Then, imagine that linear actuation wasn’t an option. You could achieve similar motion with rotary motion if you mounted one end of each linkage to an arm (rotated by a servo motor, in this case). The result is pretty similar: you’re changing the distance between linkage points, forcing the platform to move around.

And that has an interesting side effect: the “elbows” where the arms and linkages connect can dip below the plane the axes of motor rotation sit on. When the elbows can go below the robot’s body, it can walk around on them. The six elbows become six feet.

Those “legs” only have a single degree of freedom each. But when you have six, you can still achieve a walking gait. The cool part is that Stewy can still work as a Stewart platform. The “platform” is the head, which can move and tilt according to those six degrees of freedom.

Patten’s most recent video about Stewy focuses on giving the robot the ability to feel, in order to improve its performance. Specifically, Patten wanted to detect force by inferring servo torque, which he achieved by measuring current. He did that using an ESP32 development board, with a simple resistor and an extra jumper connection to an analog pin. That’s actually measuring voltage, but because the resistor value is known, he can easily calculate current. And because voltage is known, he can easily calculate power draw. And because he knows the commanded speed, he can easily calculate torque.

A custom PCB helped Patten with wiring, so he could test the hardware and see if all that theory was useful. Initially, it looked like he was getting data so noisy it was useless: a leg touching the ground wasn’t visible in the values. But eventually he realized that the data just looked like noise. Current draw would vary a lot from one millisecond to the next, but by sampling at a fast rate and averaging chunks, he could see the trends.

Now, Stewy can feel the ground and will be able to walk with a bit more sophistication.