Gilbert Documentation

Some core tenants of Gilbert (Important: this is my current assessment but is certainly up for negotiation. Contact me for suggestions)

  1. The core of Gilbert is a virtual conscious, which is capable of using multiple sewing technologies to emulate typical biologic senses (primarily sight, touch, and sound) in order to localize its own body in a world of other objects.

    1. to emphasize, the robot should have a defined virtual "world" defined by objects, not pixels, surfaces, polygons, or other basics geometries, but actual objects that are the tied to their expected renders for different sensors . (Ex: object floor0001 might have a particular texture, hardness, and reverb that can be verified with cameras and other sensors, to assert to the robot where it exists in the world). Likewise, a chair0001 might have a certain geometry, a certain instance, and certain textures that allow the robot to place it within it’s mental model / virtual conscious, which is validated by different sensors. In this way, the robot is acting and running entirely within its own simulated world that is merely being updated by external simuli, just like when we turn off the light before we go to bed, we still know how to get there because we have a full physical model of our environment in our head, even if we have lost our simuli (sight).

  2. The next tenant of Gilbert is an emphasis on soft robotics. When we grab an object, we don’t typically grab it by thinking of the precise position/ shape our fingers will take (though certainly for some objects where we grab the object and how or fingers / thumbs support it are very important ) but we are usually focused (if conscious of it at all) on the amount of force / pressure we are applying to the object. We don’t want to loose grip on the object, but for sensitive things like eggs, decorations, and some tools, we also don’t want to apply so much pressure that we would crack or deform whatever it is we are interacting with. To assist in these capabilities, a robot frame can consider a combination of 3 design options:

    1. Either a naturally or controlled deforming of the interacting layer with the object to get even pressure across the object (this could be as simple as something like foam on the fingers / palm, or as complicated as many independent articulations on the fingers to control the general position and therefore force on the different finger / hand elements )

    2. By using actuators or intermediate force transfers that dampen / act as shock observation and allow some bending to occur in response to changes in pressure, that would give less precise position control, but would also give a "ramp" in how the pressure / force changes as a response to position.

    3. By using extremely precise actuators, the position can be controlled extremely well in order to have good control of the pressure (but this approach to me seems fairly complex)