Bootstrapping Artificial Evolution to Design Robots for Autonomous Fabrication

Round 1

Reviewer 1 Report

The authors present an interesting approach for the combined evolution of physical robots and their controllers, according to the evolution of things paradigm. For the robots, authors propose using flexible 3D printing technologies, while for the controllers a CPPN-based approach is used. In this regard, the authors focus on two main problems: how to achieve that robots proposed by the EA are compatible with the manufacturing process, and how to generate the initial population in the EA. The paper is technically sound and covers an interesting topic. However, three are the main concerns that do not allow its publication in its current format:

a) The paper is unnecessarily long, which makes it being unfocused on the ultimate goal of the work. A lot of side information is provided, not necessary for the understanding of the proposal. In particular, the introduction, discussion, and concluding remark sections are difficult to follow. Moreover, the organization is not clear to me. The main proposal of the paper is in the Experimental framework section, which also makes it difficult to understand where the actual contributions of the paper are.

b) The organization of the paper, more similar to a project report than a journal, makes that within the discussion, some issues, such as the examples of manufactured robots, are provided. These are some examples since this is ongoing work. However, this also makes it difficult to understand where the limits of the proposal are. Authors state in the introduction that manufacturing in evolutionary robotics is a problem and that in most existing work only simulations are used. However, at the end of the day, in this proposal, the authors rely only on simulations again, since manufacturing the robots is mainly an ongoing work. So, the limits of the current (not future) proposal must be clearly identified.

c) It is also necessary to clearly reflect which is the progress of this work with respect to reference [|2] of the same authors. It should be clearly stated in the introduction if this is an extension of this conference paper. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The objective of this paper, entitled "Bootstrapping artificial evolution to design robots for autonomous fabrication", is to develop a method that generates feasible and diverse initial populations of robot designs.

The authors present two main contributions:
(1) a method for producing diverse manufacturable robots, and
(2) an assesment of these robots, based on their ability to solve a maze-navigation problem.

The literature review includes relevant references.
The methods are clearly presented. The study is methodologically sound.
The presented conclusions are supported by the results.
The quality of English is good, with minor typos.
Overall, it was a pleasure to read this paper.

My main concern, which is not addressed in the paper, but which authors hint at in their future work, is:
"How much shape/morphology complexity can an evolutionary algorithm sustain, with a given morphology-encoding structure, while still being capable of identifying morphologies and controllers for solving specific tasks?"
Based on my experience, the inability to do targeted optimisation of the morphology/controller makes the search for a solution quite slow.
How far can you get with artificial evolution, before you run into problems with the dimensionality of the solution space? Can you do better than artificial evolution?

Please find below my suggestions.

For the literature review, regarding methods for producing robots, you may find interesting this reference on origami robots:
[*] "Design, fabrication and control of origami robots"
Daniela Rus, Michael T Tolley
Nature Reviews Materials 3 (6), 101-112, 2018

Table 1 is not very clear.
How many wheels/sensors does a robot with 3 organs have?
In the "feature descriptor" evolution, did all the robots with 4 organs have only casters, and no wheels/sensors/joints?

In section 4.5, it would be great to see fulfilled the expectations generated in the beginning of the paper about experiments with physical robots.
The robotic assembly with the "Robot factory" system is difficult to evaluate with the presented results/photos. The paper lacks proof of the Robot Fabricator ability to assemble these robots (but this is neither critical, nor central to the subject of this paper).

Line 540: "[...] bootstrap population [...] have enough body plans to solve the environments without the need to evolve the body plan any further."
This is dependent on the task being solved, and may not be true for more complex tasks (where evolution may be necessary).

Minor remarks:
- Line 52: reproduction -> sexual reproduction
- Line 225: duplication in phrase.
- Figure 6 caption has a typo, and should be: "The spatial descriptor generates the robots with the _lowest_ number of organs [...]".
- Figure 12: please increase font size in plots' axes.
- Multiple problems with plurals in English.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The paper can be accepted now in its current form.