Programming Manufacturing Robots with Imperfect AI: LLMs as Tuning Experts for FDM Print Configuration Selection

Ekta U. Samani; Christopher G. Atkeson

cs.ROMar 23, 2026

Programming Manufacturing Robots with Imperfect AI: LLMs as Tuning Experts for FDM Print Configuration Selection

Authors: Ekta U. Samani, Christopher G. Atkeson

Organizations: Robotics Institute at Carnegie Mellon University

Abstract

We use fused deposition modeling (FDM) 3D printing as a case study of how manufacturing robots can use imperfect AI. In FDM, print configuration strongly affects output quality. Yet, novice users typically rely on default configurations, trial-and-error, or direct recommendations from generic AI models (e.g., ChatGPT). These strategies can produce complete prints, but they do not reliably meet specific objectives. We present a modular approach that treats an LLM as a source of tuning expertise. We embed this source of expertise within a Bayesian optimization loop. An approximate evaluator scores each candidate print configuration and returns structured diagnostics, which the LLM uses to propose natural-language adjustments that are compiled into machine-actionable guidance for optimization. On 100 Thingi10k parts, our LLM-guided loop achieves the best configuration on 78% objects with 0% likely-to-fail cases, while direct AI model recommendations are rarely best and exhibit 15% likely-to-fail cases. These results suggest that LLMs provide more value as constrained decision modules in optimization loops than as end-to-end oracles for print configuration selection. We expect this result to extend to broader LLM-based robot programming.

Figures & tables

Fig. 1: Overview of the modular LLM-guided optimization loop for FDM 3D print configuration selection. An evaluator computes objective value $Obj(x)$ and structured diagnostics, an LLM proposes corrective changes, and a compiler maps these changes to guidance ( $\mathcal{V}(x),\mathcal{I}$ ) used for optimization. Lighter boxes denote example intermediate outputs.

Change type (example cues)	Residual	Params
Directional preference (increase $u$ by $k$ )	$\mathrm{ReLU}(k-\Delta u)$	$k$
Equality/target (set $u=\alpha$ ) ( $u$ should equal $v$ )	$(u-\alpha)^{2}$ $(u-v)^{2}$	$\alpha$
Range/box ( $u\in[L,U]$ ; $u\geq L$ )	$\mathrm{ReLU}(L-u)$ $+\;\mathrm{ReLU}(u-U)$	$L,U$
Margin (make $u\geq v+\delta$ )	$\mathrm{ReLU}((v+\delta)-u)$	$\delta$
Ratio target ( $u/v\approx\beta$ )	$\left(u/(v+\varepsilon)-\beta\right)^{2}$	$\beta$
Sum cap ( $\sum_{i}u_{i}\leq k$ )	$\mathrm{ReLU}\!\left(\sum_{i}u_{i}-k\right)$	$k$

TABLE I: Residual templates for the LLM Guidance Compiler

Aggregator (example cues)	Form	Params
Soft-AND (conjunction): “all of …”	$1-\prod_{i}(1-\tilde{\rho}_{i})^{w_{i}}$	$w_{i}$
Soft-OR (disjunction/ at least one): “ $u$ or $v$ , any of these”	$\prod_{i}(\tilde{\rho}_{i})^{w_{i}}$	$w_{i}$

TABLE II: Violation aggregators for the LLM guidance compiler

Method	Median	% of objects that are best or within $x\%$ of best			Flagged likely to fail
	$Obj(x)$	Best	Within 1%	Within 5%	Flagged likely to fail
Original orientation w/ default parameters	0.37	0	2	12	6
Heuristic orientation w/ default parameters	0.26	3	4	15	9
ChatGPT 5.2 Thinking	0.25	8	20	50	15
Gemini 3 Pro (Chat Interface)	0.21	4	18	48	15
(Unguided) Optimization	0.24	7	14	44	0

TABLE III: End-to-end performance of print-configuration selection methods, evaluated with

Obj(x)

(Eq. 1 ).

Fig. 2: Sample efficiency of guided (ours) vs. unguided optimization. Curves show best-so-far $Obj(x)$ averaged over 10 initializations and objects. Shading shows 95% bootstrap confidence intervals over objects.

Row beats column	None	Llama 3.1 70B	GPT-5.2 API	Hand- crafted	Mean winrate
None	–	0.08	0.08	0.08	0.08
Llama 3.1 70B	0.92	–	0.46	0.50	0.63
GPT-5.2 API	0.92	0.54	–	0.54	0.67
Handcrafted	0.92	0.50	0.46	–	0.63

TABLE IV: Pairwise win-rate matrix for guidance-source ablation.

Row beats column	LLM prompted w/o examples	LLM prompted with examples	Hand- crafted	Mean winrate
LLM prompted w/o examples	–	0.38	0.39	0.38
LLM prompted with examples	0.62	–	0.50	0.56
Handcrafted	0.61	0.50	–	0.56

TABLE V: Pairwise win-rate matrix for prompting ablation (with vs. without examples). LLM is Llama 3.1 70B.

Row beats column	None	LLM: one action	LLM: two actions	Hand- crafted	Mean winrate
None	–	0.32	0.08	0.08	0.16
LLM: one action	0.68	–	0.24	0.22	0.38
LLM: two actions	0.92	0.76	–	0.50	0.72
Handcrafted	0.92	0.78	0.50	–	0.74

TABLE VI: Pairwise win-rate matrix for action-budget ablation (0/1/2 actions per iteration). LLM is Llama 3.1 70B.

Fig. 3: Qualitative comparison of toolpath previews and photos of 3D prints for three representative parts across methods. Our method uses GPT-5.2 (medium reasoning) guidance. Numbered callouts mark key print issues and configuration changes. Preview colors follow the legend.

Explore similar work

Sep 27, 2026cs.RO

Large Language Models for Model-Based Robot Design

Large Language Models (LLMs) can contribute useful engineering knowledge to robot design, but directly generated designs may rely on implicit assumptions and provide no guarantees of feasibility or optimality. These assumptions are critical because different reasonable modeling choices can materially change which designs are predicted to be feasible or optimal. We therefore present a framework that uses LLMs to construct explicit engineering models containing physical relationships, compatibility constraints, and objectives, allowing these modeling choices to be inspected and revised before formal optimization. The model can then be updated with additional engineering, manufacturer, or system-specific information before formal multi-objective optimization provides feasibility and Pareto-optimality guarantees with respect to the finalized model and specified design space. We evaluate the framework on quadcopter and line-following robot component-selection problems. Across 30 direct LLM design trials, none could be verified as feasible under the corresponding finalized model. Comparisons with an independently developed expert model and successive stages of model refinement further showed that changes in modeling assumptions substantially altered the predicted feasible and Pareto-optimal design sets. Together, these results show that using LLMs to construct explicit engineering models makes the underlying design choices available for inspection and revision before those assumptions determine the optimized designs. Explicit modeling therefore provides an interface for combining LLM-generated engineering knowledge, system-specific information, and formal design optimization.

Andrew Wilhelm, Angelina Zhao, Nils Napp

Department of Electrical and Computer Engineering, Cornell University, Ithaca, NY, 14850, United States

Engineering Optimal Design Robot Navigation

Ask PDF

May 5, 2026cs.LG

LLM-ADAM: A Generalizable LLM Agent Framework for Pre-Print Anomaly Detection in Additive Manufacturing

Additive manufacturing (AM) continues to transform modern manufacturing by enabling flexible, on-demand production of complex geometries across diverse industries. Fused filament fabrication (FFF) has extended AM to laboratories, classrooms, and small production environments, but this accessibility shifts process-planning responsibility to users who may lack manufacturing expertise. A syntactically valid slicer profile can still encode thermally or geometrically harmful settings, and subtle G-code edits can alter extrusion, cooling, or adhesion before a print begins. Pre-print G-code screening catches accidental or adversarial machine-program errors before material or machine time is wasted. This paper proposes LLM-ADAM as a generalizable LLM framework for pre-print anomaly detection in AM. The framework decomposes the task into three roles: Extractor-LLM maps a G-code file to a structured process-parameter schema; Reference-LLM converts printer and material documentation into aligned operating ranges; and Judge-LLM interprets a deterministic deviation table and G-code evidence to decide whether a part is non-defective or belongs to an anomaly class. Printers, materials, and LLM backbones are interchangeable test conditions, not fixed assumptions. We evaluate the framework on an N=200 FFF G-code corpus spanning two desktop printer families, two materials, and five classes including non-defective, under-extrusion, over-extrusion, warping, and stringing. The best framework configuration reaches 87.5% accuracy, compared with 59.5% for the strongest engineered single-LLM baseline. The results show that structured decomposition, rather than backbone strength alone, is the dominant source of improvement, with defect classes identified at or near ceiling for leading configurations while residual errors concentrate on conservative false alarms for non-defective samples.

Ahmadreza Eslaminia, Chuhan Cai, Cameron Smith +5

Department of Mechanical Science and Engineering, University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA · Coordinated Science Laboratory, University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA · Department of Mechanical Engineering, University of Michigan, Ann Arbor, MI 48109, USA +1

Graph Anomaly Detection Large Language Model Agents Slices Adam

Ask PDF

May 8, 2026cs.AI

FactoryBench: Evaluating Industrial Machine Understanding

We introduce FactoryBench, a benchmark for evaluating time-series models and LLMs on machine understanding over industrial robotic telemetry. Q&A pairs are organized along four causal levels (state, intervention, counterfactual, decision) instantiating Pearl's ladder of causation, and span five answer formats: four structured formats are scored deterministically and free-form answers are scored by an LLM-as-judge voting protocol. We propose a scalable Q&A generation framework built around structured question templates, present FactoryWave (a dense, multitask, multivariate sensor dataset collected from a UR3 cobot and a KUKA KR10 industrial arm), and construct FactoryBench as a large-scale benchmark of over 70k Q&A items grounded in roughly 15k normalized episodes from FactoryWave, AURSAD, and voraus-AD. Zero-shot evaluation of six frontier LLMs shows that no model exceeds 50% on structured levels or 18% on decision-making, revealing a wide gap between current models and operational machine understanding.

Yanis Merzouki, Coral Izquierdo, Matei Ignuta-Ciuncanu +8

1ETH Zurich · 2Forgis · 3UC3M +4

Large-Scale Benchmark Natural Language Processing

Ask PDF