pith. sign in

arxiv: 2606.02832 · v1 · pith:YJV7NVZAnew · submitted 2026-06-01 · 💻 cs.AI

An Exploration of Collision-based Enemy Morphology Generation

Johor Jara Gonzalez , Matthew Guzdial This is my paper

Pith reviewed 2026-06-28 14:15 UTC · model grok-4.3

classification 💻 cs.AI
keywords procedural content generationenemy morphologycollision informationvideo gamesevolutionary algorithmsmorphology generationgame AI
0
0 comments X

The pith

Three collision-based approaches generate video game enemy morphologies with performance matching or exceeding an evolutionary baseline from robotics.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper addresses the gap in procedural content generation for video game enemies by focusing on morphologies, or basic body plans defined by collision data. It presents three new methods that build these shapes directly from player collision information. Each method carries distinct strengths and weaknesses, yet all achieve results equivalent to or better than an evolutionary algorithm adapted from prior robotics research. A sympathetic reader would care because successful enemy morphology generation could support more varied and dynamic opponents in games without requiring hand-crafted designs.

Core claim

Three novel approaches to generating enemy morphologies based on player collision information each provide different strengths and weaknesses, but all achieve equivalent or better performance than an evolutionary baseline adapted from prior robotics morphology work.

What carries the argument

Player collision information used to construct enemy body plans and collision shapes.

If this is right

  • Designers gain multiple options for generating varied enemy types suited to different game genres.
  • Collision-based generation provides a viable alternative to evolutionary methods for morphology tasks in games.
  • The approaches support procedural creation of enemies that interact meaningfully with player movement.
  • Metrics originally developed for robotics can guide evaluation in game contexts with minimal adjustment.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • These methods could support real-time enemy adaptation during play sessions rather than pre-generation only.
  • Similar collision-driven techniques might apply to generating shapes for non-enemy elements such as platforms or items.
  • Combining the approaches with existing level-generation tools could produce more integrated game content.

Load-bearing premise

Performance metrics and evaluation setups adapted from robotics morphology generation are valid for judging whether generated enemy shapes produce playable and fun video game experiences.

What would settle it

A controlled playtest in which human players consistently rate the generated enemies as unplayable or unfun despite the morphologies scoring well on the adapted robotics metrics.

Figures

Figures reproduced from arXiv: 2606.02832 by Johor Jara Gonzalez, Matthew Guzdial.

Figure 1
Figure 1. Figure 1: Output of the enemy morphology representation from different generators along with artist concepts using these as [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Environment Representation in Unity, where the [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Full pipeline where the first enemy representation [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Enemy visualization Both GA variants show very low DO (around 0.05 on average), meaning that they rarely depart substantially from the initial ran￾dom enemies, and their diversity is also limited (average SD below 0.1). Finally, computation time strongly distinguishes the methods: A*-based Production Rules variant completes in approximately 2.4 seconds, whereas all RL and GA variants require on the order o… view at source ↗
Figure 5
Figure 5. Figure 5: Randomly selected outputs that were delivered to an artist to conceptualize 4 different sprites per output. [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Visualization of the 10 random enemies initializa [PITH_FULL_IMAGE:figures/full_fig_p012_6.png] view at source ↗
read the original abstract

Despite a great deal of prior research into Procedural Content Generation (PCG), relatively little prior work has explored generating enemies for video games. In particular, there is almost no work on generating enemy morphologies, the basic body plan or collision information for in-game enemies, despite the existence of related morphology generation work in robotics. In this paper, we explore three different novel approaches to generate enemy morphologies based on player collision information. We found that each approach provides different strengths and weaknesses, but all had equivalent or better performance than an evolutionary baseline adapted from prior robotics morphology work.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 1 minor

Summary. The manuscript claims to introduce three novel collision-based approaches for generating enemy morphologies in video games, each with distinct strengths and weaknesses, and reports that all three achieve equivalent or better performance than an evolutionary baseline adapted from prior robotics morphology work.

Significance. If the results hold under game-appropriate evaluation, the work would address a documented gap in PCG for enemy design by providing concrete generation methods and a direct comparison to an external baseline. The experimental framing is a strength, but the transfer of robotics-derived metrics requires explicit validation to establish relevance to playable game content.

major comments (1)
  1. [Evaluation / Results] The central performance claim rests on metrics and protocols adapted from robotics morphology generation. No evidence is provided that these metrics (e.g., locomotion efficiency or structural stability) correlate with game-relevant properties such as challenge balance, fairness, or engagement. This directly affects whether the reported superiority supports the paper's goal of generating playable enemies.
minor comments (1)
  1. [Abstract] The abstract states the performance finding at a high level but supplies no information on the three approaches, the exact metrics, sample sizes, or statistical tests, limiting immediate assessment of the result.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their constructive feedback. We address the major comment on evaluation metrics below and outline a targeted revision to improve clarity without altering the core contributions.

read point-by-point responses

  1. Referee: [Evaluation / Results] The central performance claim rests on metrics and protocols adapted from robotics morphology generation. No evidence is provided that these metrics (e.g., locomotion efficiency or structural stability) correlate with game-relevant properties such as challenge balance, fairness, or engagement. This directly affects whether the reported superiority supports the paper's goal of generating playable enemies.

    Authors: We agree that the manuscript does not provide direct evidence correlating the robotics-derived metrics (locomotion efficiency, structural stability) with game-specific properties such as challenge balance, fairness, or engagement. The evaluation protocol was chosen to enable a consistent, apples-to-apples comparison with the adapted evolutionary baseline from prior robotics work. To address this gap, we will add a dedicated paragraph in the Discussion section acknowledging the transfer limitations of these metrics to playable game content and explicitly noting that future validation through human-subject playtesting would be required to establish relevance to engagement and balance. This revision clarifies the scope of the performance claims without changing any reported results. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical comparison to external robotics baseline

full rationale

The paper reports an experimental study generating enemy morphologies via three collision-based methods and directly compares their performance metrics against an evolutionary baseline adapted from prior (non-author-overlapping) robotics work. No derivation chain reduces a result to its own inputs by definition, no fitted parameters are relabeled as predictions, and no load-bearing claims rest on self-citations or imported uniqueness theorems. The central claim is an empirical ranking under stated metrics; any concern about metric suitability for games is an external-validity issue, not circularity.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only; no free parameters, axioms, or invented entities are described.

pith-pipeline@v0.9.1-grok · 5608 in / 925 out tokens · 19327 ms · 2026-06-28T14:15:18.524589+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

41 extracted references · 4 canonical work pages · 1 internal anchor

  1. [1]

    Eric Butler, Kristin Siu, and Alexander Zook. 2017. Program synthesis as a generative method. InProceedings of the 12th International Conference on the Foundations of Digital Games. 1–10

    2017

  2. [2]

    Raúl Lara Cabrera, Mariela Nogueira Collazo, Carlos Cotta Porras, and Antonio José Fernández Leiva. 2015. Procedural content generation for real-time strategy games.IJIMAI3, 2 (2015), 40–48

    2015

  3. [3]

    Michael Cook and Simon Colton. 2014. A rogue dream: Automatically generating meaningful content for games. InProceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, Vol. 10. 2–7

    2014

  4. [4]

    Michael Cook, Simon Colton, Azalea Raad, and Jeremy Gow. 2013. Mechanic miner: Reflection-driven game mechanic discovery and level design. InEuropean Conference on the Applications of Evolutionary Computation. Springer, 284–293

    2013

  5. [5]

    Sam Earle, Maria Edwards, Ahmed Khalifa, Philip Bontrager, and Julian Togelius

  6. [6]

    In2021 IEEE Conference on Games (CoG)

    Learning controllable content generators. In2021 IEEE Conference on Games (CoG). IEEE, 1–9

  7. [7]

    Sam Earle, Zehua Jiang, and Julian Togelius. 2024. Scaling, control and general- ization in reinforcement learning level generators. In2024 IEEE Conference on Games (CoG). IEEE, 1–8

    2024

  8. [8]

    Roberto Gallotta, Graham Todd, Marvin Zammit, Sam Earle, Antonios Liapis, Julian Togelius, and Georgios N Yannakakis. 2024. Large language models and games: A survey and roadmap.IEEE Transactions on Games(2024)

    2024

  9. [9]

    Johor Jara Gonzalez, Seth Cooper, and Matthew Guzdial. 2023. Mechanic Maker 2.0: reinforcement learning for evaluating generated rules. InProceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, Vol. 19. 266–275

    2023

  10. [10]

    Pablo Gutiérrez-Sánchez, Marco A Gómez-Martín, Pedro A González-Calero, and Pedro P Gómez-Martín. 2021. Reinforcement learning methods to evaluate the impact of AI changes in game design. InProceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, Vol. 17. 10–17

    2021

  11. [11]

    2025.Procedural content generation via machine learning: An Overview

    Matthew Guzdial, Sam Snodgrass, and Adam J Summerville. 2025.Procedural content generation via machine learning: An Overview. Springer

    2025

  12. [12]

    David Ha. 2019. Reinforcement learning for improving agent design.Artificial life25, 4 (2019), 352–365

    2019

  13. [13]

    Emily Halina and Matthew Guzdial. 2022. Diversity-based Deep Reinforcement Learning Towards Multidimensional Difficulty for Fighting Game AI.arXiv preprint arXiv:2211.02759(2022)

    work page arXiv 2022

  14. [14]

    Hart, Nils J

    Peter E. Hart, Nils J. Nilsson, and Bertram Raphael. 1968. A Formal Basis for the Heuristic Determination of Minimum Cost Paths.IEEE Transactions on Systems Science and CyberneticsSSC-4, 2 (1968), 100–107

    1968

  15. [15]

    Yomna MI Hassan and Amal Aboulhassan. 2024. A Framework for Adaptive Game Play and Adversary Creation Based on Gamer Behavioral Analysis and Large Language (LLM) Models. In2024 International Mobile, Intelligent, and Ubiquitous Computing Conference (MIUCC). IEEE, 413–417

    2024

  16. [16]

    Mark Hendrikx, Sebastiaan Meijer, Joeri Van Der Velden, and Alexandru Iosup

  17. [17]

    Procedural content generation for games: A survey.ACM Transactions on Multimedia Computing, Communications, and Applications (TOMM)9, 1 (2013), 1–22

    2013

  18. [18]

    Nicholas Jennings and Björn Hartmann. 2024. Towards Runtime Behavior Gen- eration in Games. (2024)

    2024

  19. [19]

    Arthur Juliani, Vincent-Pierre Berges, Ervin Teng, Andrew Cohen, Jonathan Harper, Chris Elion, Chris Goy, Yuan Gao, Hunter Henry, Marwan Mattar, and Danny Lange. 2020. Unity: A general platform for intelligent agents.arXiv preprint arXiv:1809.02627(2020). https://arxiv.org/pdf/1809.02627.pdf

    work page arXiv 2020

  20. [20]

    Joel Lehman and Kenneth O Stanley. 2011. Evolving a diversity of virtual creatures through novelty search and local competition. InProceedings of the 13th annual conference on Genetic and evolutionary computation. 211–218

    2011

  21. [21]

    Mahdi Farrokhi Maleki and Richard Zhao. 2024. Procedural content generation in games: A survey with insights on emerging llm integration. InProceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, Vol. 20. 167–178

    2024

  22. [22]

    Sudhanshu Maurya, Arnav Kotiyal, Pankaj Prusty, Nilesh Shelke, Abhishek Bhat- tacherjee, and Tiyas Sarkar. 2025. Optimizing NPC Behavior in Video Games Using Unity ML-Agents: A Reinforcement Learning-Based Approach. In2025 3rd International Conference on Disruptive Technologies (ICDT). IEEE, 1601–1606

    2025

  23. [23]

    Kathryn Merrick and Mary Lou Maher. 2006. Motivated reinforcement learning for non-player characters in persistent computer game worlds. InProceedings of the 2006 ACM SIGCHI international conference on Advances in computer entertain- ment technology. 3–es. An Exploration of Collision-based Enemy Morphology Generation Conference acronym ’XX, June 03–05, 20...

    2006

  24. [24]

    Tim Nämerforslund. 2021. Machine Learning Adversaries in Video Games: Using reinforcement learning in the Unity Engine to create compelling enemy characters

    2021

  25. [25]

    Leonardo T Pereira, Breno MF Viana, and Claudio FM Toledo. 2021. Procedural enemy generation through parallel evolutionary algorithm. In2021 20th Brazilian Symposium on Computer Games and Digital Entertainment (SBGames). IEEE, 126–135

    2021

  26. [26]

    David Ponton, Yannick Francillette, Hugo Tremblay, and Bruno Bouchard. 2025. Analyzing Enemy Behavior Difficulty: An Automated Framework Exploiting Behavior Trees. In2025 IEEE Gaming, Entertainment, and Media Conference (GEM). IEEE, 1–6

    2025

  27. [27]

    John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov

  28. [28]

    Proximal policy optimization algorithms.arXiv preprint arXiv:1707.06347 (2017)

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  29. [29]

    Karl Sims. 2023. Evolving virtual creatures. InSeminal Graphics Papers: Pushing the Boundaries, Volume 2. 699–706

    2023

  30. [30]

    Junru Song, Yang Yang, Wei Peng, Weien Zhou, Feifei Wang, and Wen Yao. 2024. MorphVAE: advancing morphological design of voxel-based soft robots with variational autoencoders. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 38. 10368–10376

    2024

  31. [31]

    Kynan Sorochan and Matthew Guzdial. 2022. Generating real-time strategy game units using search-based procedural content generation and monte carlo tree search.arXiv preprint arXiv:2212.03387(2022)

    work page arXiv 2022

  32. [32]

    Megan Sumner, Vardan Saini, and Matthew Guzdial. 2024. Mechanic maker: accessible game development via symbolic learning program synthesis. InPro- ceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, Vol. 20. 235–244

    2024

  33. [33]

    Julian Togelius, Noor Shaker, and Mark J Nelson. 2014. Procedural content generation in games: A textbook and an overview of current research.Togelius N. Shaker M. Nelson Berlin: Springer1 (2014)

    2014

  34. [34]

    Julian Togelius, Georgios N Yannakakis, Kenneth O Stanley, and Cameron Browne

  35. [35]

    InEuropean Conference on the Applications of Evolutionary Computation

    Search-based procedural content generation. InEuropean Conference on the Applications of Evolutionary Computation. Springer, 141–150

  36. [36]

    Andrew Wood and Adam Summerville. 2019. Understanding Boss Battles: A Case Study of Cuphead.California State Polytech. Univ., San Luis Obispo, CA, USA, Tech. Rep(2019)

    2019

  37. [37]

    Chao Yu, Akash Velu, Eugene Vinitsky, Jiaxuan Gao, Yu Wang, Alexandre Bayen, and Yi Wu. 2022. The surprising effectiveness of ppo in cooperative multi-agent games.Advances in neural information processing systems35 (2022), 24611–24624

    2022

  38. [38]

    Yuzhong Zhang, Guixuan Zhang, and Xinyuan Huang. 2022. A survey of proce- dural content generation for games. In2022 International Conference on Culture- Oriented Science and Technology (CoST). IEEE, 186–190

    2022

  39. [39]

    Mohammad Zohaib. 2018. Dynamic difficulty adjustment (DDA) in computer games: A review.Advances in Human-Computer Interaction2018, 1 (2018), 5681652

    2018

  40. [40]

    Alexander Zook and Mark Riedl. 2014. Automatic game design via mechanic generation. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 28

    2014

  41. [41]

    Alexander Zook and Mark O Riedl. 2014. Generating and adapting game me- chanics. InProceedings of the 2014 Foundations of Digital Games Workshop on Procedural Content Generation in Games. A Random Enemies Initialization Visualization This appendix reports the randomly generated initial enemy set used to initialize all experimental runs. The intent of incl...

    2014