Hi @Alex, I’m curious about the in-between solutions. For example, if I pick up a pencil, I can indirectly get a “pretty OK” sensation through it. Of course, it does not match my own limbs, but being mechanically coupled to the pencil allows my skin, etc., to feel the vibrations and kinematics from the pencil indirectly. This came up when I was listening to a podcast about people’s proficiency with prosthetics without a direct neural interface. Are you familiar with any intermediate approaches like this, where there is a high concentration of high-fidelity sensors mechanically coupled to end effectors to indirectly infer sensations?
Hi @tslominski, I am not aware of any such technology available off the shelf. This would be the whisker approach. I guess it can work well for sensing a texture but not so great for pressure. A good skin technology would do both. Nature only uses whiskers in limted ways, primarily I think for animals to ensure a gap is wide enough for their body to get through. I have considered fitting whiskets to my robot but there doesn’t seem to be anything out there you can buy. If I were to design one I guess I would use either a piezoelectric device or magnietic coil at the base of the whisker, much like a vinyl record player stylus. In fact that might be a good place to start for sensing texture if you don’t need long whiskers, move a stylus across a surface and digitise the output. That only gives you one point of contact, having a small array of whiskers would be mechanically tricky but could be done.
Ah, fair, it is the whisker approach. What I have in mind, though, is usually outside of what people think of as whiskers. The setup I have in mind here is if you were walking or rolling on your “whiskers”, much like a human can walk on stilts or prosthetics, drive a car, or ride a bike. There is a nonzero amount of information we interpret as occurring at the end of the end effector, even though there are no sensors there. All available because our highly sensitive sensors end up mechanically coupled to remote end effectors.
But where is the advantage over putting the sensors on the end effector? Unless you spread a small, high resolution sensor over a wider area perhaps. For optical sensors which are high resolution but over a tiny area, expanding the GelSight technique using many conical push rods.
Please note that I’m speaking from ignorance. I am only loosely aware that some sensors are too big, too fragile, too expensive, too something to end up on the end effector. My thinking here is to consider a design space on a (hypothetical) Pareto frontier where a large, fragile sensor in the main body might work for a use case, and perhaps this configuration would be better than having no sensor or using a “less good” sensor at the end effector.
High-precision CNC touch probes also come to mind, it’s a ruby-tipped “whisker”:
Electric motors and magnetic actuators can also perform rudimentary “end-effector” sensing via current draw and back-EMF voltage. That’s how washing machine load sensing works.
(Back-EMF voltage can also be used to estimate motor shaft angle, via a technique called “flux observer”. Although, from practical experience, it is easier said than done…)
