LCM

c. 2012

Low-Cost Manipulator

A major focus of our R&D work at Meka was trying to find novel ways to get past using expensive harmonic drives that almost all high-performance human-scale robots were using at the time. We put our energy into finding ways to leverage the geometry of the arm and try to utilize the space between joints to support really strange transmission designs. Most of our early developments used cables and pulley configurations that ran the whole length of an arm segment as the transmission. Some of the major benefits were significantly lower cost materials and manufacturing (at least for the individual elements), greatly reduced transmission inertia and it concentrated the mass (e.g. the motor) up the arm toward the base which helped increase the payload and speed.

The second major focus was on finding ways to squeeze performance from cheap motors to get away from using expensive and heavy Maxon motors. The first main challenge was that these types of motors were used for quadcopters and didn’t have any sensors for feedback of the rotor position. This isn’t a problem for quadcopter designs because running them with open-loop control is totally feasible with the more advanced designs using back-EMF as a feedback for more efficient control. The goal of a quadcopter motor controller is to get the motor to a high-velocity and keep it there. For robotics, this is almost the opposite need. We needed good control at and near zero velocity to be able to perform dexterous manipulation tasks. We initially attempted to embed hall-effect sensors near the windings to measure the magnet positions - the standard approach of most industrial motors for using trapezoidal commutation. But this lead to a couple of issues:

1. It was hard to retrofit these motors and we were way too small to get manufacturers to install the sensors as well as the manufacturer just not set up to do it. It would have been way too much of an adventure for our timeline.

2. The off-the-shelf motor controllers typically used in industrial robotics were not able to handle the very low inductance typical of these quadcopter motors.

This drove us to learn how to build our own motor controllers which would let us leverage the cost and torque density of the motors and generally give us more flexibility in control and packaging for our robot. I developed a simple way to use a single optical encoder on the motor output to commutate the motor without needing hall-effect sensors embedded in the motor. This “phase-lock” process only happened once very quickly at startup and let us integrate and control any BLDC motor with full field-oriented control. To handle very low inductance motors I pushed the switching and motor control to 100 kHz so we could get good current measurements and tight current-control loops.

The LCM was an interesting experiment focused on combining 3 new ideas for designing robot arms: cable drives, bent sheet-metal structures and very low-cost motors for quadcopters. We learned a lot and eventually kept iterating and honing the motor controllers for very low inductance motors and cable drive in future machines. We largely abandoned developing designs that used welding - way too hard to keep the tolerances and wouldn’t really scale well but sheet metal plates and structure would continue in a few more designs.