Real-time Learning and Evolution in Robotic Art Installations
cs.HC · cs.LG · cs.NE · cs.RO
Quick verdict
Practice-based account of three learning robotic artworks; rich design insight, but audience-effect and emergence claims rest on anecdote, not evaluation.
Summary
The authors present three robotic art installations — Vessels (2015), Open-Ended Ensemble (2016), and Morphosis (2025) — in which machine learning and evolutionary computation run in real time as aesthetic material rather than optimization tools. Vessels couples behavior trees with a custom genetic algorithm over binary 'personality' genomes that recombine when 10–20 water robots meet; Open-Ended Ensemble uses competitive coevolution via genetic programming, pitting a population of probe-control programs against one controlling fluorescent bulbs; Morphosis applies tabular Q-learning with tile coding so three silicone-skinned robots learn simple behaviors (e.g., 'Rocking without rolling', intrinsic-curiosity 'Being curious') within a 15–20 minute performance. Framed as research-creation, the paper argues the works instantiate rather than simulate Darwinian evolution, expose adaptation on two timescales (lifetime learning and evolution), and reshape artist and audience roles through concepts like Umwelt and apprivoisement. Practical design lessons cover presentation contexts, timescale compression, and material channels (red/green reward lighting, legend-free visualization). The works also fed back into published engineering research (Kelly & Heywood 2018; Kelly et al. 2023). Main limitations: no empirical audience evaluation — claims of empathy and narrative projection rest on anecdote (footnote 1, one visitor quote); algorithmic parameters are described only at high level; and non-reproducibility is embraced as inherent, limiting scientific scrutiny.
Review complete
Assessment at a glance
4
risk flags identified
● high 0
● medium 2
● low 2
5 strengths · 5 weaknesses · 5 author questions
AI text signal = Binoculars zero-shot detector score over 20 sampled passages (Fast-DetectGPT co-reported) — absence does not prove human authorship, and a flag does not imply misconduct.
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- The three installations use learning and evolutionary processes as aesthetic experience in itself, not as a means to optimize a specific solution (Sections 1, 2, Conclusion).
- The works are instantiations, not simulations, of Darwinian evolution — actual variation, selection, and inheritance on populations of digital organisms (Section 1, after Pennock 2007).
- Real-time embodied learning resists simulation by design, mandating in-situ adaptation and artist residencies (Section 3.3).
- Research-creation and Artificial Life share a synthetic epistemology in which building systems is a legitimate mode of knowledge production (Section 3.4).
- Presentation context, timescale design, and material choices determine whether adaptive processes become legible to audiences (Section 4).
Strengths
Three complementary case studies with real exhibition histories
Vessels, Open-Ended Ensemble, and Morphosis span behavior trees + GA, coevolutionary GP, and tabular Q-learning (Sections 2.1–2.3), with documented presentations in galleries, museums, and public space (Figures 1–4, 7–9) — grounding the discussion in working systems, not proposals.
Documented two-way art-to-science transfer
Open-Ended Ensemble's genetic programming extended into peer-reviewed evolutionary RL (Kelly & Heywood 2018) and robot control (Kelly et al. 2023); Morphosis' ML fed Audry et al. 2020 (Section 3.4) — the strongest verifiable evidence that the practice generates engineering knowledge.
Analytically useful two-timescale framework
Concrete audience and material design practices
Clear conceptual positioning
Weaknesses
No empirical audience evaluation
Algorithmic underspecification
Central behavioral claims undemonstrated
Non-reproducibility embraced, leaving claims unfalsifiable
Editorial inconsistencies
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
Report preview
Scope & Claims
This is a research-creation paper by Sofian Audry (UQAM) and Stephen Kelly (McMaster) describing three robotic art installations built across a decade: **Vessels** (2015; 10–20 water robots; behavior trees plus a custom genetic algorithm over binary 'personality' genomes that recombine on encounter, Section 2.1), **Open-Ended Ensemble** (2016; competitive coevolution via genetic programming, pitting a probe-controlling program population against one controlling fluorescent bulbs, Section 2.2), and **Morphosis** (2025; three silicone-skinned spheroid robots running tabular Q-learning with tile coding through a 15–20 minute performance of named behaviors such as 'Rocking without rolling' and intrinsic-curiosity 'Being curious', Section 2.3). The central claims are explicitly non-optimization claims: learning and evolution here function as aesthetic material rather than solvers (Sections 1, 2, Conclusion); the works are 'instantiations' rather than simulations of Darwinian evolution (Section 1, after Pennock 2007); adaptation is exposed to audiences on two timescales — lifetime learning and evolution (Figure 6, Section 4.2); and research-creation shares Artificial Life's synthetic epistemology, in which building systems is a mode of knowledge production (Section 3.4).
The remaining sections cover Methodology Assessment, Statistical Evidence, Results & Reproducibility, Limitations & Risks, and Verdict & Recommendations.
- All 5 strengths & 5 weaknesses, analyzed in detail
- 4 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
What are the three robotic artworks described in the paper?
Vessels (2015): a swarm of 10–20 water robots whose 'personalities' evolve via a genetic algorithm layered on behavior trees; Open-Ended Ensemble (2016): robotic probes and fluorescent lights locked in competitive coevolution driven by genetic programming; Morphosis (2025): three silicone-skinned spheroid robots learning simple behaviors with tabular Q-learning during a 15–20 minute gallery performance.
Do the robots in these installations solve tasks or optimize performance?
No. The authors deliberately use learning and evolution not to optimize solutions but as an aesthetic experience: hesitant, clumsy, or failed behaviors are treated as part of the artwork's value, and the search process itself is what audiences witness in real time.
Is code or data available for these installations?
No public code, data, or hardware documentation is released. The paper states that reproducing specific behaviors is nearly impossible because the systems learn in changing physical environments — a constraint the authors treat as inherent to the art practice.
What evidence supports the paper's claims about audience reactions?
Only anecdotal observation: artists and docents reported visitors narrating social relationships (friendship, courtship, enmity) among the Vessels robots, and one visitor of Open-Ended Ensemble described loving and hating the work simultaneously. No systematic audience study is reported.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- Text inside figures (e.g., Figure 6's evolution/learning loop labels) is extracted out of layout order; figure internals were read only via captions.
- The three Vimeo video links cannot be viewed in this review; installation behavior was assessed from text and captions only.
- The abstract renders 'machine leaning', likely an original typo though a PDF-extraction artifact cannot be fully excluded.
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.