Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

No-brainer: Morphological Computation driven Adaptive Behavior in Soft Robots

Alican Mertan, Nick Cheney

TL;DR

The paper investigates whether intelligent, adaptive behavior can emerge from morphology alone, without a dedicated brain or controller. It uses voxel-based soft robots in the EvoGym environment, with stimulus-responsive materials encoded by CPPNs and evolved via MAP-Elites to produce morphologies that adapt to binary environmental cues. The key contributions are the demonstration of closed-loop morphological computation enabling adaptive locomotion and the extension to a swarm capable of implementing logic-gate-like behaviors and memory-like dynamics. This approach suggests a path toward robust, controller-free adaptation in soft robotics and informs future work on integrating morphological computation with brain-like controllers.

Abstract

It is prevalent in contemporary AI and robotics to separately postulate a brain modeled by neural networks and employ it to learn intelligent and adaptive behavior. While this method has worked very well for many types of tasks, it isn't the only type of intelligence that exists in nature. In this work, we study the ways in which intelligent behavior can be created without a separate and explicit brain for robot control, but rather solely as a result of the computation occurring within the physical body of a robot. Specifically, we show that adaptive and complex behavior can be created in voxel-based virtual soft robots by using simple reactive materials that actively change the shape of the robot, and thus its behavior, under different environmental cues. We demonstrate a proof of concept for the idea of closed-loop morphological computation, and show that in our implementation, it enables behavior mimicking logic gates, enabling us to demonstrate how such behaviors may be combined to build up more complex collective behaviors.

No-brainer: Morphological Computation driven Adaptive Behavior in Soft Robots

TL;DR

The paper investigates whether intelligent, adaptive behavior can emerge from morphology alone, without a dedicated brain or controller. It uses voxel-based soft robots in the EvoGym environment, with stimulus-responsive materials encoded by CPPNs and evolved via MAP-Elites to produce morphologies that adapt to binary environmental cues. The key contributions are the demonstration of closed-loop morphological computation enabling adaptive locomotion and the extension to a swarm capable of implementing logic-gate-like behaviors and memory-like dynamics. This approach suggests a path toward robust, controller-free adaptation in soft robotics and informs future work on integrating morphological computation with brain-like controllers.

Abstract

It is prevalent in contemporary AI and robotics to separately postulate a brain modeled by neural networks and employ it to learn intelligent and adaptive behavior. While this method has worked very well for many types of tasks, it isn't the only type of intelligence that exists in nature. In this work, we study the ways in which intelligent behavior can be created without a separate and explicit brain for robot control, but rather solely as a result of the computation occurring within the physical body of a robot. Specifically, we show that adaptive and complex behavior can be created in voxel-based virtual soft robots by using simple reactive materials that actively change the shape of the robot, and thus its behavior, under different environmental cues. We demonstrate a proof of concept for the idea of closed-loop morphological computation, and show that in our implementation, it enables behavior mimicking logic gates, enabling us to demonstrate how such behaviors may be combined to build up more complex collective behaviors.
Paper Structure (7 sections, 5 figures)

This paper contains 7 sections, 5 figures.

Table of Contents

  1. Introduction and Background
  2. Methodology
  3. Responding to Binary Stimuli
  4. Making a Swarm for More Complex Behavior
  5. Discussion
  6. Conclusion
  7. Acknowledgements

Figures (5)

  • Figure 1: Images from the simulator environment bhatia2021evolution that shows the behavior of sensory voxels. Voxels in (a) and (b) respond to stimulus 1 and voxels in (c) and (d) respond to stimulus 2. Voxels in (a) and (c) shrink to their minimum width in response to the absence of their respective stimulus (left), and expand to their maximum width in response to the presence of their respective stimulus (right). Voxels in (b) and (d) respond the opposite way. These materials enable robots (e) that actively undergo shape change to produce new behaviors according to cues from some sensory stimuli.
  • Figure 2: The spacetime diagram of a run champion (left) and its gait under different stimuli patterns as snapshots from the simulation (right). In the spacetime diagram, the solid black line in the middle keeps track of the robot's center of mass over time. The two accompanying lines show the current stimuli pattern over time. The robot moves toward the positive x-direction when one stimulus is present, otherwise, it moves in the opposite direction. The different gait behaviors are achieved by the morphological changes caused by sensory voxels -- an example of morphological computation and how the body could be the source of adaptive behavior. See more robots in action at https://github.com/mertan-a/no-brainer/tree/master/robots-in-action
  • Figure 3: The spacetime diagrams of all run champions from independent runs, plotted together. We omit the behaviors where the robot doesn't respond to any stimulus (LLLL and RRRR behaviors). Time is normalized to reflect actuation cycles and position is normalized to reflect the robot's body length. The numbers in the bottom right corner show the average distance traveled by run champions.
  • Figure 4: Performances of run champions from 10 independent runs with a $10\times10$ bounding box. Symmetric robot behaviors are grouped by the pattern of their movement direction across stimuli (e.g. LRLR and RLRL are both represented as XYXY). Colors represent the number of stimuli a robot's morphology adapts to. Each data point is plotted and mean values are marked with an 'x'. Behaviors are separated with empty columns based on statistically significant performance differences (at $P < 0.001$) -- each distribution in a group is statistically significantly better compared to each other distribution belonging to any group to their right. Bounding boxes of size $5\times5$ and $7\times7$ are qualitatively consistent.
  • Figure 5: Top row High-level schematic of the swarm (left) and its inner wiring (right). Lines connecting to robots from the left represent stimuli and the ones on the right represent outputs. We assume that the behavior of one robot could determine the stimulus of another. Bottom row The behavior of each robot with respect to their immediate stimuli (left) and the behavior of the starred robot with respect to swarm stimuli (right). While each robot successfully performs the behavior that it evolved for with respect to its immediate stimuli, examination of the selected robot with respect to swarm stimuli shows complex behavior that would otherwise require memory to exhibit.