Boston Dynamics ATLAS robot (image credit: Boston Dynamics)

Vacuum Driven Artificial Muscles for the Future of Robotics

Artificial muscles are one of the most meaningful inventions we can come up. They’re something that would let us real android walking moving androids. A real humanoid robot whose muscular system is almost identical to ours.

Many roboticists seem to be content with the present state of robot locomotion namely pneumatics and linear actuators, but these are largely weak or bulky inelegant methods.

Vacuum driven contraction is a stronger and quieter way to move. With a compartmentalize strand a strong vacuum could compress the compartments and contract the strand. As the pockets partially contract with the vacuum it brings the two anchor points closer together just like our muscles.

This approach could employ hundreds of strands to exert a much stronger force on the whole and a fine amount of force depending on the strength of the vacuum.

My design for the compartmentalized strand. The compartments contract when the air is sucked out.

The biggest drawback to this approach is wiring the individual strands. There would need to be tubes that suck the air out of the strands to contract the compartments. Furthermore this tube would need to have a firm reliable seal.

Relaxed (left) and contracted (right) strands acting on hinge joint.

This artificial muscle would be an ideal application of flexible 3d printing. With silicone 3d printing the vacuum tube and compartmentalized strand can be made into one part so there is less likelihood of the tube separating from the strand.

Once the artificial muscle is made all that is really needed to be done is attach the artificial muscles to a skeletal frame just like our own muscle groups are attached to our bones. Once everything is attached and the wired up to the vacuum tube controller the programming can be done by a machine.

Programming more complicated movement such as walking could be done with machine learning. By putting in the parameters of the skeleton into a simulation the machine can began learning how to walk (or at least get to the objective.) Once it has done this in the simulation it can run the most successful tests on the real wired up skeleton. After a few million tests on various types of motions it would need to have in its repertoire the machine would be able to accomplish simple tasks and continue to iterate and learn, but that’s the beauty of machine learning. We just need to make the hardware, the simulation, and tasks for the program to try and solve. The machine will then train itself to accomplish those tasks.

But first we need to figure out the hardware. Then the algorithms can take care of the bulk of the movement programming for us.




Product Manager and designer writing about ideas. Living and working in SF. See more of my projects at

Love podcasts or audiobooks? Learn on the go with our new app.

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
Alex Cox

Alex Cox

Product Manager and designer writing about ideas. Living and working in SF. See more of my projects at

More from Medium

[Reinforcement Learning] TD Actor-Critic Application: Post Indicator Valve Optimization Problem

The Future of Deep Learning: Photonics

Information Technology Trends in India

Artificial Intelligence documentaries you should watch