Evolutionary Synthesis of Sensing Controllers for Voxel-based Soft Robots

Abstract

Soft robots allow for interesting morphological and behavioral designs because they exhibit more degrees of freedom than robots composed of rigid parts. In particular, voxel-based soft robots (VSRs)-aggregations of elastic cubic building blocks-have attracted the interest of Robotics and Artificial Life researchers. VSRs can be controlled by changing the volume of individual blocks: simple, yet effective controllers that do not exploit the feedback of the environment, have been automatically designed by means of Evolutionary Algorithms (EAs).In this work we explore the possibility of evolving sensing controllers in the form of artificial neural networks: we hence allow the robot to sense the environment in which it moves. Although the search space for a sensing controller is larger than its non-sensing counterpart, we show that effective sensing controllers can be evolved which realize interesting locomotion behaviors. We also experimentally investigate the impact of the VSR morphology on the effectiveness of the search and verify that the sensing controllers are indeed able to exploit their sensing ability for better solving the locomotion task.

Criticality-Driven Evolution of Adaptable Morphologies of Voxel-Based Soft-Robots

“…The controller determines how the area of each voxel varies over the time, possibly based on the readings of the sensors the VSR is equipped with. The ability of sensing both itself and the environment makes VSRs potentially more effective in performing tasks where perception is beneficial as, e.g., locomotion on uneven terrains (Talamini et al, 2019).…”

“…The second controller, that we call neural controller, is based on a fully connected feedforward artificial neural network (NN) and resembles that proposed by Talamini et al (2019). At each time step, the controller collects the vector x (k) ∈ R m of m readings from the VSR sensors and feeds it to the NN; instantaneously, the NN outputs a vector y (k) ∈ [−1, 1] n .…”

“…The control signal of the ith voxel at time is the ith element of y (k) . Based on the experimental settings of (Talamini et al, 2019), we here worked with a NN with one hidden layer with 65% of the neurons of the input layer and with the hyperbolic tangent as the activation function. An instance of the neural controller for a VSR with n voxels and m sensors is hence completely described by the vector θ ∈ R p of weights of the NN, with p 0.65(m + 1)m + 0.65mn, where the +1 is the bias.…”

“…A VSR is defined by a body (or morphology), which is simply an aggregate of voxels, and a brain, which is the control algorithm responsible for actuating its body. Biological inspired meta-heuristics such as Evolutionary Computation (EC) have been extensively applied to the domain of VSRs (Hiller and Lipson, 2011;Cheney et al, 2013;Cheney et al, 2015;Talamini et al, 2019;Medvet et al, 2020a;Ferigo et al, 2021) and have been shown to be a promising approach for both the design and control of VSRs. However the optimization of these robots is often oriented toward a specific task, e.g., a locomotion task, which provides no guarantee on the effectiveness of the resulting designs when subjected to a different task.…”

“…Mechanisms of body adaptation are less understood both for robots and biological organisms. In addition to the bodies evolved toward criticality, we have considered several other bodies for the validation process, some of them inspired by previous works such as Talamini et al (2019), while others automatically generated thorough algorithms based on randomness. We compare all the robots on the different tasks, and we draw an overall ranking that shows that criticality is indeed a good predictor of the adaptability of a robot morphology.…”