LivePhys: Transforming Static Physics Problems into Interactive Simulations via a Scan-to-Play Framework

Juan Wu; Xiangwen Zhang; Xiaoming Chen; Xiaowei Dai; Yonghong Ke; Ziyu Luo

REVIEW 4 major objections 4 minor 25 references

LivePhys: Transforming Static Physics Problems into Interactive Simulations via a Scan-to-Play Framework

T0 review · 4 major / 4 minor · reviewed 2026-08-01 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A scan-to-play pipeline turns static textbook physics diagrams into interactive simulations.

desk verdict A clean scan-to-play pipeline that clearly beats general MLLMs on layout and executability, but the key claim of physical correctness is never evaluated. read the letter →

arxiv 2607.20990 v1 pith:SCJL2QYP submitted 2026-07-23 cs.ET

Xiaowei Dai , Ziyu Luo , Xiangwen Zhang , Xiaoming Chen , Juan Wu , Yonghong Ke This is my paper

classification cs.ET

keywords physicseducationinteractivesimulationscan-to-playmultimodalperceptionintermediaterepresentationcognitiveloadtextbookdiagramsengine

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to establish that a 'scan-to-play' pipeline can reliably turn a static mechanics problem—a diagram plus a short text—into an executable, interactive simulation without manual programming. The key move is to separate perception (reading the diagram and text) from physics reasoning (deciding what the objects and constraints are) and pass the result to a deterministic physics engine. If the framework works as claimed, learners can manipulate parameters such as initial velocity or friction and watch the dynamics respond, which the authors argue reduces the cognitive load of mentally simulating motion. Evaluation on fifty textbook problems reports large gains over end-to-end multimodal-model generation in executability, spatial layout accuracy, and interaction fidelity, and a twenty-person user study reports lower perceived workload with the simulations than with static materials.

What carries the argument

The load-bearing object is the structured intermediate representation (IR): a formal description of physical entities, their properties, and inter-body constraints, produced by a multimodal language model under a step-by-step reasoning prompt and then executed by a deterministic 2D physics engine. The IR acts as the contract between perception and simulation: it forces the model to commit to concrete entities (blocks, slopes, springs), parameters (masses, coefficients, velocities), and constraints (contact, support, boundaries) before any dynamics are computed. Around this IR, the pipeline uses cross-modal grounding to bind textual labels to image regions, and symbolic-variable handling that

What would settle it

Apply LivePhys to a problem with a known analytic solution—say a block launched up an incline with a stated friction coefficient—and compare the simulated motion against the closed-form trajectory over several seconds; if the simulated position and velocity deviate beyond a small tolerance, the framework's physical reasoning is not trustworthy.

Watch

Extended reading notes

Core claim

The central claim is that decoupling multimodal perception from physics-aware reasoning makes automatic simulation generation reliable. LivePhys first segments the diagram, extracts text, and aligns textual labels to visual objects; then a multimodal language model, prompted to reason step by step, infers entities, parameters, and constraints and emits a structured intermediate representation; finally a deterministic 2D physics engine executes that representation to produce a simulation whose layout matches the original diagram and whose parameters are exposed as interactive controls. The authors report that this decomposition fixes the main failure mode of end-to-end generation: spatially i

Load-bearing premise

The load-bearing premise is that the multimodal language model infers the correct physical parameters and constraints from the parsed diagram and text; the pipeline never independently checks whether the dynamics it runs match the problem's intended physics.

Editorial extensions

If this is right

If the framework is right, the bottleneck for generating interactive simulations from the large back-catalog of textbook problems disappears: no manual authoring or programming is needed per problem.
Learners can test hypotheses in real time by adjusting symbolic variables, which the authors tie to inquiry-based learning.
Automated evaluation of generated simulations becomes possible for layout and interactivity, though not yet for physical correctness.
General-purpose multimodal models, when asked to produce code end-to-end, generate executables that run but often misplace objects; the decomposition into IR plus deterministic engine is the claimed remedy.
The user-study results, if confirmed with larger samples, suggest interactive simulation is a viable way to reduce split-attention and cognitive load in mechanics instruction.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

A natural next step the paper does not take is to verify physical correctness dynamically: compare the simulated trajectory against the analytic solution of the textbook problem, and add a checker that rejects IRs whose dynamics diverge.
The same perception-then-IR pipeline could generalize outside mechanics, e.g., to geometric optics or electrical circuits, wherever textbook diagrams encode a small set of entities and constraints.
Because the framework inherits the MLLM's physical commonsense, its failures will be silent: the simulation will run and look plausible while being physically wrong, which argues for coupling the IR with a symbolic solver as a validation layer.
The cognitive-load result, measured only by perceived workload on twenty participants, should be read as preliminary; a stronger test would measure learning outcomes, not just self-report.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

You should read this paper if you care about education-tech systems that use LLMs to generate interactive simulations. The takeaway is simple: the pipeline is thoughtfully designed and the results look encouraging, but the headline claim—that LivePhys produces physically grounded simulations—is not actually tested. Nothing in the evaluation checks whether the dynamics are physically correct.

What's new: The scan-to-play idea itself isn't new—Augmented Physics and the authors' own PhysGest are doing adjacent things. The contribution is the architecture: decoupling perception (SAM+OCR+grounding) from reasoning (MLLM) and execution (Matter.js), plus the structured intermediate representation. That's a clean, sensible design, and the execution-based evaluation is a reasonable step beyond qualitative demos. The paper also reports a user study, small but a real attempt to address cognitive load.

Where it's soft: The three metrics—executability rate, spatial accuracy, interaction fidelity—don't tell you whether the simulation is physically meaningful. A simulation can run, look like the diagram, and have sliders while the block slides uphill. The MLLM infers masses, friction, constraints, and there is no verification step. That's load-bearing because the stated purpose is helping learners form accurate mental models. So the 94%/92%/90% numbers are about code generation and layout, not physics correctness. That gap is the main weakness. Also, the baselines got image-only prompts while LivePhys got text+diagram; that's an unfair comparison, and the author-scored binary metrics could use independent or blinded scoring. The user study is preliminary, as the authors admit. No code or data released, so I can't check the pipeline myself.

Do I believe the central claim? Directionally, yes: a structured pipeline with explicit object masks and coordinate mapping should beat zero-shot MLLMs on spatial layout. But the paper overstates it as "physically consistent." A revision that adds a physics correctness check—e.g., comparing simulated trajectories against analytic solutions or human expert ratings—and releases the dataset would make this a solid contribution.

Bottom line: This is a serious systems paper that deserves referee time. The methodology needs work before publication, but the framework is coherent and likely useful. I'd send it to a venue that handles HCI or AI-for-education, and ask the authors to fix the evaluation gap. I wouldn't cite it in my own work until that's done.

Referee Report

4 major / 4 minor

Summary. The paper introduces LivePhys, a three-stage Scan-to-Play framework that converts static 2D mechanics textbook problems (diagram + text) into interactive Matter.js simulations. Stage S1 performs OCR, segmentation, and cross-modal grounding; Stage S2 uses an MLLM (GPT-4o) with chain-of-thought prompting to extract explicit/implicit/symbolic parameters and build a structured intermediate representation; Stage S3 instantiates the IR in Matter.js with interactive controls. The authors evaluate on 50 problems against zero-shot GPT-4o, Claude 3.5 Sonnet, and Gemini 1.5 Pro, reporting higher Executability Rate, Spatial Accuracy, and Interaction Fidelity, and a 20-participant NASA-TLX user study suggesting lower perceived cognitive load than static materials.

Significance. If the claims are validated, LivePhys would be a practically useful contribution to automated physics simulation generation for education. The decoupling of perception, reasoning, and deterministic execution is a sensible architecture, and the use of a structured IR with explicit constraints is a step beyond end-to-end code generation. The paper also ships a concrete 50-problem benchmark, which is valuable for future work. However, the evaluation as reported does not verify that the generated simulations are physically correct; the metrics only check executability, initial layout, and UI binding. Because the stated pedagogical goal depends on physical fidelity, this is a central gap that must be addressed before the outperformance claim can be accepted.

major comments (4)

[Section IV-B, metrics (ER/SA/IF)] The central claim of 'physically consistent' simulations is not tested. The three metrics—Executability Rate, Spatial Accuracy, and Interaction Fidelity—only check that code runs, the initial layout resembles the diagram, and sliders are bound. They do not validate that MLLM-inferred masses, friction coefficients, initial velocities, or constraint relations (Section III-B) match the problem statement. A simulation with a block sliding up a plane, a 2 kg mass treated as 200 kg, or a non-conserving collision passes all three metrics. Since the paper's purpose is to help learners form accurate mental models, a physically wrong simulation is a pedagogical hazard. Please add a physical-correctness evaluation on a subset (or all) of the 50 problems: for example, compare simulated trajectories, accelerations, final velocities, or equilibrium conditions against analytic solutions or standard tex
[Section IV-B, baseline prompt] The baseline comparison is unfair because the prompt given to GPT-4o, Claude 3.5 Sonnet, and Gemini 1.5 Pro is 'Write a complete Matter.js simulation code based on this image. Ensure the layout matches the diagram and include interactive sliders.' Baselines receive only the image, while LivePhys receives both the image and the accompanying problem text. Since many physical parameters (masses, friction coefficients, initial velocities) appear in the text rather than the image, the baselines are starved of information critical to physics correctness. Please run the baselines with a matched prompt that includes the transcribed problem text (e.g., 'based on this image and the problem text: ...'), or justify why text is excluded. Without this, the reported superiority over general-purpose MLLMs may reflect input asymmetry rather than framework superiority.
[Section IV-B, Spatial Accuracy / Interaction Fidelity] SA and IF are binary scores assigned by 'two authors independently, with disagreements resolved by discussion.' No inter-rater reliability statistic is reported, and the authors are not blind to the method being scored (LivePhys vs. baselines). Because these metrics are central to the main comparison, the scoring procedure is vulnerable to expectation bias. Please report Cohen's kappa or another inter-rater reliability measure, and ideally use an independent annotator who is blind to the source method. In addition, the differences in Table I are not accompanied by any statistical significance test; with n=50, a McNemar's test or bootstrap confidence intervals would be appropriate.
[Section IV-C, cognitive load user study] The second central claim—that LivePhys-generated simulations reduce learners' perceived cognitive load—rests on a small study (n=20, 10 per group) using only self-reported NASA-TLX scores. There is no objective learning outcome measure, no comparison to an alternative interactive tool (so the effect might be due to interactivity per se, not LivePhys's quality), and the paper itself acknowledges the tripartite cognitive load framework limitation. While the large effect sizes (d≈1.1–1.5) are suggestive, the abstract's 'demonstrates' overstates the evidence. Please either soften the claim to 'provides preliminary evidence' in the abstract and conclusions, or add a post-intervention problem-solving test and (ideally) an active control condition to substantiate the pedagogical benefit.

minor comments (4)

[Abstract / Table I] The abstract says 'significantly outperforms' but Table I reports no significance tests. Provide at least p-values or confidence intervals for the 50-problem comparison, or replace 'significantly' with 'consistently.'
[Section III-B.1] The implicit-parameter mapping rules (e.g., 'smooth surface'→μ=0, 'free fall'→a=g) are only sketched. Include the full mapping list in an appendix or supplementary material; otherwise the reader cannot assess coverage or potential bias in the reasoning stage.
[Section IV-C] Report the t-statistics and degrees of freedom for the reported differences, not just p-values and effect sizes. Also state whether a Bonferroni or other multiple-comparison correction was applied to the three TLX dimensions.
[Section V] The conclusion says 'physically grounded' simulations, but the evaluation does not demonstrate physical grounding. Please align the wording with the evidence (e.g., 'spatially consistent and executable simulations subject to the correctness of MLLM-inferred parameters').

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper is an empirical system comparison; no parameter is fitted to the evaluation and the only self-citation is not load-bearing.

full rationale

LivePhys makes no mathematical derivation that could reduce to its own inputs. The pipeline is an engineering system: multimodal perception (OCR/SAM), MLLM-based intermediate representation generation with explicit Chain-of-Thought prompting and hand-coded domain rules, and deterministic Matter.js execution. The reported gains over GPT-4o, Claude 3.5 Sonnet, and Gemini 1.5 Pro are empirical comparisons against external zero-shot baselines on executability, spatial accuracy, and interaction fidelity. No parameter is fitted to the evaluation set, and no output metric is defined in terms of the pipeline's own outputs in a way that forces the result; indeed Spatial Accuracy is 92%, not 100%, so the measured outcome is not guaranteed by construction. The only self-citation, PhysGest [17], appears in related work as a point of contrast and is not load-bearing for any central claim. The hand-coded mapping rules (e.g., 'smooth surface' to mu=0) bake domain assumptions into the prompt, but they are stated inputs, not fitted predictions, so they do not constitute circularity. The acknowledged limitations (small user study, NASA-TLX only) are external-validity and measurement concerns, not circular-reasoning concerns.

Assumptions & free parameters 1 free parameters · 6 assumptions · 0 invented entities

No numerical parameters are fitted in the traditional sense; the system relies on a set of hand-specified domain mapping rules and off-the-shelf components. The load-bearing assumptions are listed.

free parameters (1)

Implicit-parameter mapping rules = e.g., smooth surface → μ=0; free fall → g=9.8
Hand-specified domain rules in the S2 prompt (Section III-B1) that translate natural-language phrases into concrete physical values; chosen by the authors, not learned or validated against data.

assumptions (6)

domain assumption All physical entities in the diagram are segmentable by SAM from edge-detection prompts.
S1 relies on SAM producing candidate regions that correspond to physical bodies; failure (e.g., overlapping or sketched objects) would break the pipeline.
domain assumption The MLLM reasoning controller can correctly infer entities, parameters, and constraints from the parsed IR and problem text.
S2's entire output depends on this; no verification step checks the inferred constraints against the intended physics.
domain assumption NASA-TLX perceived workload is a valid proxy for learning-relevant cognitive load.
The user study's conclusion rests on this; the authors themselves note it does not directly map to the tripartite cognitive load framework.
domain assumption Matter.js deterministically models the textbook mechanics scenarios in scope.
The final simulation's physical behavior is taken as correct by execution; the paper does not validate trajectory or collision dynamics against analytic solutions.
ad hoc to paper The hand-coded implicit-parameter mapping rules are valid.
Section III-B1 embeds these rules in the prompt; they are not learned or validated, but they are standard textbook conventions.
domain assumption The 50 collected problems are representative of textbook mechanics rigid-body problems.
Selection could bias toward diagram structures the pipeline handles well.

how reviews work

0 comments

Cite this review

Pith. "Pith review of LivePhys: Transforming Static Physics Problems into Interactive Simulations via a Scan-to-Play Framework." pith.science (2026). https://pith.science/paper/SCJL2QYP

@misc{pith2026260720990,
  author       = {Pith},
  title        = {Pith review of: LivePhys: Transforming Static Physics Problems into Interactive Simulations via a Scan-to-Play Framework},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SCJL2QYP}},
  note         = {Machine review of arXiv:2607.20990}
}

read the original abstract

Physics problems in textbooks are typically presented as static diagrams accompanied by brief textual descriptions, requiring learners to infer dynamic physical behaviors through mental visualization. This process often imposes high cognitive demands and limits learners' ability to form accurate mental models. In this paper, we present \textbf{LivePhys}, a framework that enables a \emph{Scan-to-Play} paradigm for mechanics learning by transforming static textbook physics problems into executable, interactive simulations. LivePhys decouples multimodal perception from physics-aware reasoning and deterministic simulation. Given a problem diagram and its accompanying text, LivePhys performs text extraction, geometric segmentation, and cross-modal grounding to construct a structured, physics-aware intermediate representation. A multimodal large language model is then used as a reasoning controller to infer entities, parameters, and constraints, which are executed by a physics engine to generate spatially consistent and interactive simulations that allow learners to explore and manipulate problem conditions dynamically. Our evaluation results show that LivePhys significantly outperforms general-purpose multimodal models in simulation executability, spatial accuracy, and interaction fidelity. In addition, a user study demonstrates that interacting with LivePhys-generated simulations reduces learners' perceived cognitive load compared to static textbook materials.

Figures

Figures reproduced from arXiv: 2607.20990 by the authors.

**Figure 2.** Illustration of the architecture of LivePhys. LivePhys automates a [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Interface of a complete generated simulation. It includes the parsed [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗

Figures from the paper (2 more)

**Figure 4.** Figure 4: Examples of LivePhys-generated simulations with visual scaffolding across different mechanics scenarios: (a) incline dynamics with force labels; (b) [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]

**Figure 5.** Figure 5: Comparison of cognitive load based on NASA-TLX scores. Partici [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

25 extracted references · 4 linked inside Pith

[1]

Cognitive load during problem solving: Effects on learning,

J. Sweller, “Cognitive load during problem solving: Effects on learning,” Cognitive Science, vol. 12, no. 2, pp. 257–285, 1988

1988
[2]

Managing split-attention and redundancy in multimedia instruction,

S. Kalyuga, P. Chandler, and J. Sweller, “Managing split-attention and redundancy in multimedia instruction,”Applied Cognitive Psychology: The Official Journal of the Society for Applied Research in Memory and Cognition, vol. 13, no. 4, pp. 351–371, 1999

1999
[3]

Scientific discovery learning with computer simulations of conceptual domains,

T. De Jong and W. R. Van Joolingen, “Scientific discovery learning with computer simulations of conceptual domains,”Review of educational research, vol. 68, no. 2, pp. 179–201, 1998

1998
[4]

Videophy: Evaluating physical commonsense for video generation,

H. Bansal, Z. Lin, T. Xie, Z. Zong, M. Yarom, Y . Bitton, C. Jiang, Y . Sun, K.-W. Chang, and A. Grover, “Videophy: Evaluating physical commonsense for video generation,”arXiv preprint arXiv:2406.03520, 2024

arXiv 2024
[5]

Matter.js: A 2d rigid body physics engine for the web,

L. Brummitt, “Matter.js: A 2d rigid body physics engine for the web,” https://github.com/liabru/matter-js, 2014

2014
[6]

Interactive multimodal learning environ- ments: Special issue on interactive learning environments: Contemporary issues and trends,

R. Moreno and R. Mayer, “Interactive multimodal learning environ- ments: Special issue on interactive learning environments: Contemporary issues and trends,”Educational psychology review, vol. 19, no. 3, pp. 309–326, 2007

2007
[7]

Phet: Interactive simulations for teaching and learning physics,

K. Perkins, W. Adams, M. Dubson, N. Finkelstein, S. Reid, C. Wieman, and R. LeMaster, “Phet: Interactive simulations for teaching and learning physics,”The physics teacher, vol. 44, no. 1, pp. 18–23, 2006

2006
[8]

Algodoo: A tool for encouraging creativity in physics teaching and learning,

B. Gregorcic and M. Bodin, “Algodoo: A tool for encouraging creativity in physics teaching and learning,”The Physics Teacher, vol. 55, no. 1, pp. 25–28, 2017

2017

Show all 25 references

[9]

ophysics,

T. Walsh, “ophysics,” https://ophysics.com/, 2025, accessed: August 23, 2025

2025
[10]

Augmented physics: Creating interactive and embedded physics simula- tions from static textbook diagrams,

A. Gunturu, Y . Wen, N. Zhang, J. Thundathil, R. H. Kazi, and R. Suzuki, “Augmented physics: Creating interactive and embedded physics simula- tions from static textbook diagrams,” inProceedings of the 37th Annual ACM Symposium on User Interface Software and Technology, 2024, pp. 1–12

2024
[11]

Phet: Simulations that enhance learning,

C. E. Wieman, W. K. Adams, and K. K. Perkins, “Phet: Simulations that enhance learning,”Science, vol. 322, no. 5902, pp. 682–683, 2008

2008
[12]

Learn to explain: Multimodal reasoning via thought chains for science question answering,

P. Lu, S. Mishra, T. Xia, L. Qiu, K.-W. Chang, S.-C. Zhu, O. Tafjord, P. Clark, and A. Kalyan, “Learn to explain: Multimodal reasoning via thought chains for science question answering,”Advances in Neural Information Processing Systems, vol. 35, pp. 2507–2521, 2022

2022
[13]

Gpt-4 technical report,

J. Achiam, S. Adler, S. Agarwal, L. Ahmad, I. Akkaya, F. L. Aleman, D. Almeida, J. Altenschmidt, S. Altman, S. Anadkatet al., “Gpt-4 technical report,”arXiv preprint arXiv:2303.08774, 2023

2023 arXiv
[14]

Visual instruction tuning,

H. Liu, C. Li, Q. Wu, and Y . J. Lee, “Visual instruction tuning,” Advances in neural information processing systems, vol. 36, pp. 34 892– 34 916, 2023

2023
[15]

Sora: A review on background, technology, limitations, and opportunities of large vision models,

Y . Liu, K. Zhang, Y . Li, Z. Yan, C. Gao, R. Chen, Z. Yuan, Y . Huang, H. Sun, J. Gaoet al., “Sora: A review on background, technology, limitations, and opportunities of large vision models,”arXiv preprint arXiv:2402.17177, 2024

2024 arXiv
[16]

Video diffusion models,

J. Ho, T. Salimans, A. Gritsenko, W. Chan, M. Norouzi, and D. J. Fleet, “Video diffusion models,”Advances in neural information processing systems, vol. 35, pp. 8633–8646, 2022

2022
[17]

Physgest: Transforming static textbook diagrams into physically realistic videos for enhanced physics education,

X. Dai, X. Zhang, Q. Zhang, Z. Z. Hu, B. Qian, and X. Chen, “Physgest: Transforming static textbook diagrams into physically realistic videos for enhanced physics education,” inSIGGRAPH Asia 2025 Educator’s Forum, 2025, pp. 1–6

2025
[18]

A computational approach to edge detection,

J. Canny, “A computational approach to edge detection,”IEEE Transac- tions on pattern analysis and machine intelligence, no. 6, pp. 679–698, 1986

1986
[19]

Segment anything,

A. Kirillov, E. Mintun, N. Ravi, H. Mao, C. Rolland, L. Gustafson, T. Xiao, S. Whitehead, A. C. Berg, W.-Y . Loet al., “Segment anything,” inProceedings of the IEEE/CVF international conference on computer vision, 2023, pp. 4015–4026

2023
[20]

Chain-of-thought prompting elicits reasoning in large language models,

J. Wei, X. Wang, D. Schuurmans, M. Bosma, F. Xia, E. Chi, Q. V . Le, D. Zhouet al., “Chain-of-thought prompting elicits reasoning in large language models,”Advances in neural information processing systems, vol. 35, pp. 24 824–24 837, 2022

2022
[21]

30, 2026

OpenAI, “GPT-4o,” https://openai.com/index/gpt-4o-system-card/, 2024, accessed: Apr. 30, 2026

2024
[22]

Claude 3.5 Sonnet,

Anthropic, “Claude 3.5 Sonnet,” https://www.anthropic.com/news/claude- 3-5-sonnet, 2024, accessed: Apr. 30, 2026

2024
[23]

Gemini 1.5 Pro,

Google, “Gemini 1.5 Pro,” https://gemini.google.com/app, 2024, ac- cessed: Apr. 30, 2026

2024
[24]

Evaluating large language models trained on code,

M. Chen, J. Tworek, H. Jun, Q. Yuan, H. P. D. O. Pinto, J. Kaplan, H. Edwards, Y . Burda, N. Joseph, G. Brockmanet al., “Evaluating large language models trained on code,”arXiv preprint arXiv:2107.03374, 2021

2021 arXiv
[25]

Development of nasa-tlx (task load index): Results of empirical and theoretical research,

S. G. Hart and L. E. Staveland, “Development of nasa-tlx (task load index): Results of empirical and theoretical research,” inAdvances in psychology. Elsevier, 1988, vol. 52, pp. 139–183

1988

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.

[1] [1]

Cognitive load during problem solving: Effects on learning,

J. Sweller, “Cognitive load during problem solving: Effects on learning,” Cognitive Science, vol. 12, no. 2, pp. 257–285, 1988

1988

[2] [2]

Managing split-attention and redundancy in multimedia instruction,

S. Kalyuga, P. Chandler, and J. Sweller, “Managing split-attention and redundancy in multimedia instruction,”Applied Cognitive Psychology: The Official Journal of the Society for Applied Research in Memory and Cognition, vol. 13, no. 4, pp. 351–371, 1999

1999

[3] [3]

Scientific discovery learning with computer simulations of conceptual domains,

T. De Jong and W. R. Van Joolingen, “Scientific discovery learning with computer simulations of conceptual domains,”Review of educational research, vol. 68, no. 2, pp. 179–201, 1998

1998

[4] [4]

Videophy: Evaluating physical commonsense for video generation,

H. Bansal, Z. Lin, T. Xie, Z. Zong, M. Yarom, Y . Bitton, C. Jiang, Y . Sun, K.-W. Chang, and A. Grover, “Videophy: Evaluating physical commonsense for video generation,”arXiv preprint arXiv:2406.03520, 2024

arXiv 2024

[5] [5]

Matter.js: A 2d rigid body physics engine for the web,

L. Brummitt, “Matter.js: A 2d rigid body physics engine for the web,” https://github.com/liabru/matter-js, 2014

2014

[6] [6]

Interactive multimodal learning environ- ments: Special issue on interactive learning environments: Contemporary issues and trends,

R. Moreno and R. Mayer, “Interactive multimodal learning environ- ments: Special issue on interactive learning environments: Contemporary issues and trends,”Educational psychology review, vol. 19, no. 3, pp. 309–326, 2007

2007

[7] [7]

Phet: Interactive simulations for teaching and learning physics,

K. Perkins, W. Adams, M. Dubson, N. Finkelstein, S. Reid, C. Wieman, and R. LeMaster, “Phet: Interactive simulations for teaching and learning physics,”The physics teacher, vol. 44, no. 1, pp. 18–23, 2006

2006

[8] [8]

Algodoo: A tool for encouraging creativity in physics teaching and learning,

B. Gregorcic and M. Bodin, “Algodoo: A tool for encouraging creativity in physics teaching and learning,”The Physics Teacher, vol. 55, no. 1, pp. 25–28, 2017

2017