The problem with the author's vision is that geared brushless servo actuators like the Gyems they used, all generally similar to the original MIT Cheetah actuat

The problem with the author's vision is that geared brushless servo actuators like the Gyems they used, all generally similar to the original MIT Cheetah actuators (or steppers, another popular motor) require precision low-backlash gearboxes with large bearings and servos with high-speed control electronics. They also need high-resolution force/torque sensors to be collaborative capable, and if you require it to not fall down when E-stopped, you need brakes. All of that has become impressively cheap for what it is, but scale and progress can only do so much - that hardware will always be expensive. The problem is they're starting this effort from the basic joint being an electric motor. When you start there, this all follows as what you have to do to build a robot, you're forced into a corner in your design. Instead, it's far better to have the safety and operation come inherently from the way the product is designed. I saw an impressive demo at an automation conference about a decade ago, it was a cobot where the motion was based on inflation of pneumatic bladders. It functioned similar to the way biological muscles work. It did not need gear reduction, or force sensors, or brakes, or high-speed amplifiers. If it ran into something (or something ran into it), it didn't strip gearboxes, the bladder just complied. I think that will be the 'killer cobot' someday, but it's going to be hard to impossible to launch into the industrial space.