I reviewed a couple of papers
- Physical reservoir computing with origami and its application to robotic crawling
- Recent advances in physical reservoir computing: A review
Reservoir Computing: Main takeaways (rough)
- Physical body of robot takes part in performing low-level control tasks
- Morphological computation, such as the physical reservoir computing (PRC), wherein a physical body performs genuine computations.
- It is based on artificial RNNs- It is suitable for temporal information processing
- In RC, the neural network (aka. the “reservoir”) has fixed interconnections and input weights, and only the linear output readout weights are trained by simple techniques such as, linear or ridge regression.
- The reservoir’s fixed nature opens up the possibility of using physical bodies—such as a random network of nonlinear spring and mass oscillators , tensegrity structures , and soft robotic arms —to conduct computation,
- More importantly, robotic bodies with sufficient nonlinear dynamics can also perform like a physical reservoir and directly generate locomotion gait without using the traditional controllers
- In RC, the input data is transformed into spatiotemporal patterns in a high-dimensional space by an RNN in the reservoir. Then a pattern analysis from the spatiotemporal patterns is done in the readout
- The main characteristic of RC is that the input weights (Win) and the weights of the recurrent connections within the reservoir (W) are not trained whereas only the readout weights (Wout) are trained with a simple learning algorithm such as linear regression. This simple and fast training process makes it possible to drastically reduce the computational cost of learning compared with standard RNNs, which is the major advantage of RC
- The role of the reservoir in RC is to nonlinearly transform sequential inputs into a high-dimensional space such that the features of the inputs can be efficiently read out by a simple learning algorithm.
- A motivation for physical implementation of reservoirs is to realize fast information processing devices with low learning cost.
- Origami can act as a reservoir and even simplest of origami pattern can be turned into peristaltic crawling robot powered by reservoir computing.
-
- A subset of the creases are designated as input creases and a subset ( or all of them) can be regarded as output creases
- They were able to show an origami pattern achieve a crawling gait with this method
The first paper was able to demonstrate that origami can be used for RC, the same way it has been demonstrated in soft robots, tensegrity structures and other physical phenomenon.
This is rough outline, i will update with more details later.