The scale of the problem is staggering. The Argo programme, which has deployed more than 4,000 profiling floats across the world’s oceans, returns upward of 100,000 temperature and salinity profiles every year, forming the backbone of the Global Ocean Observing System. Along the United States coastline, roughly 90 moored and drifting buoys operated by the National Data Buoy Center complement this network with sub-hourly surface measurements. Together, these systems feed the data streams that hurricane trackers, sea-level researchers, and climate modellers depend on every day. But calibration drift — a slow, systematic creep in sensor readings away from true environmental values — is easy to miss in real time and expensive to ignore in hindsight.

Drift originates from several distinct physical processes. Biofouling deposits on conductivity cell membranes gradually skew salinity readings. Pressure transducer membranes fatigue under repeated compression cycles. Optical sensors lose sensitivity as anti-fouling coatings degrade, and electronic components shift their operating characteristics as they age in the corrosive deep-ocean environment. Each affected float can corrupt months or even years of otherwise sound observations before the problem is caught. Uncorrected salinity drift in particular propagates into ocean heat content estimates, sea-level reconstructions, and Atlantic Meridional Overturning Circulation transport calculations — errors that become nearly impossible to remove once they have cycled through data assimilation systems.

Established correction methods, most notably the Owens–Wong–Cabanes approach, compare each float’s profile against the World Ocean Atlas climatology using salinity–temperature relationships to estimate sensor bias. While scientifically sound, the method is reactive, manual, and slow by design. A trained oceanographer must examine diagnostic plots, judge whether an anomaly reflects genuine instrument drift or natural water-mass variability, and apply a correction only after accumulating enough profiles — a process that routinely takes three to six months after drift onset. Fewer than 55 percent of Argo profiles are currently in delayed mode, meaning the majority of data in operational use has not received expert quality control at all. With an expanding fleet of biogeochemical Argo sensors measuring oxygen, pH, and nitrate, the human resource bottleneck is already severe and growing worse.

DriftGuard attacks the problem with a five-component architecture that reads like a blueprint for autonomous scientific infrastructure. At its core sits a physically-informed heterogeneous sensor graph spanning 135 nodes drawn from three real marine datasets, with edges weighted by haversine distance and oceanographic proximity. Sensors are connected only when they fall within an 800-kilometre basin-scale threshold, chosen because temperature–salinity covariance remains statistically meaningful across that range in the North Atlantic. The centrepiece is a novel Hybrid Temporal Transformer and Graph Attention Network with 212,481 parameters — the first such combination applied to marine sensor calibration drift detection. The Transformer captures slow within-sensor drift trajectories over weeks and months, while the graph attention network resolves ambiguity by comparing each sensor against its physically coherent neighbours.

The logic behind this two-stage design mirrors the physics of drift itself. Drift in ocean sensors is fundamentally a temporal phenomenon, developing gradually through biofouling and membrane fatigue. Yet natural ocean variability — water mass movements, mesoscale eddies, frontal crossings — can produce signals in an isolated sensor that closely resemble drift. If nearby floats in the same water mass agree with each other but disagree with one sensor, that disagreement is strong evidence of instrument fault rather than genuine oceanographic change. A novel cross-sensor consistency loss, which penalises spatial disagreement between neighbouring sensors during training, forces the encoder to learn from network-wide consensus rather than isolated signals alone.

What separates DriftGuard from every published predecessor is what happens after detection. Each anomaly score passes through a conformal prediction wrapper that produces distribution-free, statistically valid uncertainty certificates — the first rigorous marginal miscoverage guarantee in any marine sensor drift detection pipeline, bounding the false positive rate at 10 percent on future exchangeable data. Flagged sensors then enter a three-stage agentic loop orchestrated through LangGraph: Diagnose, Act, and Verify. The system retrieves relevant knowledge from a store of 2,183 indexed chunks built from 580 pages of real oceanographic quality-control documentation, then selects among five corrective tools: offset correction, statistical recalibration, satellite crosscheck, human review, or quarantine. A correction is accepted only if the post-correction consistency score falls below the conformal threshold, and verified corrections are routed back to the encoder for continuous self-improvement — the first closed retraining loop of its kind in marine sensor calibration.

The results are striking. Evaluated on Argo float profiles from 116 floats in the North Atlantic between 2021 and 2024, NDBC buoy time series from 15 stations spanning 2019 to 2024, and the expert-labelled AdapAD benchmark, DriftGuard achieved a graph-level ROC-AUC of 0.8376 and a zero-shot AUC of 0.8423 on AdapAD data it had never seen during training. Most remarkably, it detects gradual drift types an average of 84.7 readings earlier than the operational Argo QC Test 16 baseline — which, at roughly 6-hourly profile intervals, translates to approximately three weeks of reduced data contamination per drifting float. Across a fleet of more than 4,000 active floats, autonomous early detection at this level could recover hundreds of float-years of clean oceanographic data annually.

In an operational fleet evaluation across 131 sensors, the system’s predictive module identified 13 floats in a CRITICAL state requiring correction within 24 hours, with 27 more classified as HIGH risk and 38 as MEDIUM. All ten highest-risk sensors were Argo floats concentrated in the North Atlantic, and five carried seasonal flags indicating their risk elevation follows predictable annual cycles the system can anticipate in advance. The prescriptive module translated these forecasts into prioritised, resource-aware action recommendations with urgency timelines attached, while two agent-verified HIGH-tier sensors had already been corrected autonomously — reducing the manual workload for senior oceanographers. The agentic pipeline processed all four injected drift event types with a 100 percent verification success rate, and when the researchers replaced their deterministic test model with a real large language model, it produced physically grounded, drift-type-specific diagnoses and action selections consistent with Argo quality control best practices.

The system is not without limits. DriftGuard performs least reliably on sensors at the outer edges of the graph with fewer than five neighbours, where cross-sensor consensus signals are weak — 14 of 23 false negatives came from sensors with fewer than 10 connections, a concern for undersampled regions such as the South Atlantic and Indian Ocean. Threshold-based rules also retain a marginal advantage for abrupt step drift, where instantaneous discontinuities favour simple spike detection. Yet the ablation study tells a compelling story: a rule-based Z-score baseline achieved a near-random AUC of 0.4378, unsupervised machine learning raised it to 0.7421, a deep network without graph information reached 0.7885, and the full hybrid system climbed to 0.8376 — isolating the specific contribution of graph structure. As the Argo programme expands toward biogeochemical sensors and Deep Argo floats profiling to 6,000 metres, the researchers argue that the future of ocean stewardship lies not in waiting for experts to catch what machines miss, but in systems that catch, correct, and learn from drift before the record itself is compromised.

Subject of Research: Autonomous AI-based detection and correction of calibration drift in distributed marine sensor networks

Article Title: DriftGuard: an agentic hybrid temporal transformer and graph attention network framework for autonomous calibration drift detection and correction in distributed marine sensor networks

Article References: DriftGuard: an agentic hybrid temporal transformer and graph attention network framework for autonomous calibration drift detection and correction in distributed marine sensor networks. (n.d.). https://doi.org/10.3389/fmars.2026.1977051

Image Credits: AI Generated

DOI: 10.3389/fmars.2026.1977051

Keywords: DriftGuard, Argo floats, calibration drift, marine sensor networks, graph neural networks, temporal transformer, conformal prediction, agentic AI, LangGraph, ocean observation, anomaly detection, Frontiers in Marine Science

The system, described in the journal Discover Artificial Intelligence, focuses on the anamnesis, the structured history-taking phase of a medical encounter that can supply up to 80 percent of the information needed for a diagnosis. Students choose from five clinical cases, including type 2 diabetes, chronic obstructive pulmonary disease, and hypothyroidism, and receive no hint of the chief complaint before they begin, mirroring the blank-slate conditions of a genuine first consultation. They can type questions or dictate them through a browser speech-recognition interface, and the simulated patient replies with text accompanied by synthesized speech.

What distinguishes the platform from a simple chatbot is its architecture. The interview is modeled as a directed graph in which each node performs one processing stage, implemented with the LangGraph framework, and a shared state object persists across every turn. Because language models are stateless between invocations and long interviews exceed any context window, the team added a semantic memory layer built with the LangMem library. Clinical facts disclosed by the patient are stored as vector embeddings in a PostgreSQL database extended with the pgvector extension, and retrieved by cosine similarity, so the agent can return to earlier topics using different wording without contradicting itself. Each query retrieves the three most relevant memories, a value tuned empirically to keep response times under two seconds.

Realism comes from a second processing layer that transforms medically accurate replies into the voice of one of five personality profiles: know-it-all, unfriendly, friendly, forgetful elderly, or skeptical. The researchers used probabilistic prompt chaining and few-shot prompting to reshape the output, then synthesized it into audio with dynamically selected voices. The result is an encounter that behaves less like a decision tree and more like a person with moods, inconsistencies, and a memory of what has already been said.

The grading engine is equally ambitious. The team reviewed six validated communication assessment instruments, including the Communication Assessment Tool and the Communication Skills Scale, distilled 88 candidate items down to 24 criteria, and organized them into six categories: overall performance, interest in the patient, empathy, clarity of communication, open communication, and completeness. Each category is scored by a dedicated prompt running in an asynchronous parallel pipeline, producing a 1-to-10 score with a textual justification. Completeness follows a two-phase check in which a generated summary of the interview is first compared against the reference clinical case, then verified against the raw transcript, a correction introduced after physicians complained the tool penalized them for questions they had actually asked.

The validation study involved 21 participants: nine eighth-semester medical students, eleven practising physicians, and one healthcare technology expert, all conducting sessions in Spanish. Students rated the system 82.3 out of 100 on the Chatbot Usability Questionnaire, while the physician and expert cohort gave it 96.5, yielding a pooled usability score of 90.4. Acceptance of the automated feedback averaged 4.50 out of 5 among students and 4.21 among physicians. Interestingly, physicians rated the tool significantly more favorably than students, a gap the authors attribute to several untested hypotheses, including students’ higher baseline expectations shaped by daily exposure to general-purpose AI assistants, and their frustration with missing features such as laboratory results.

The agreement between machine and human judgment was more sobering. In an exploratory comparison against eleven physician ratings of the same interviews, rank agreement was highest for interest in the patient, with a Spearman coefficient of 0.64, and lowest for completeness, at 0.09. No dimension reached statistical significance after correction for six comparisons, and the confidence intervals were wide enough to accommodate anything from negligible to near-perfect agreement. Notably, the automated scores fell below the physician scores in every dimension, suggesting a systematic calibration offset rather than a failure to detect good communication.

The empathy result is particularly revealing. Participants in both cohorts felt the empathy scores were too low, yet empathy actually showed the second-highest rank agreement with human raters and the only confidence interval excluding zero. The system, in other words, ordered students on empathy much as physicians did while assigning harsher absolute scores. The authors point to plausible mechanical explanations: the rubric was derived from instruments covering the entire consultation, including phases where encouragement is expected, and the scoring prompts lacked anchored descriptions of what each point on the scale represents. Both are fixable without changing the underlying model.

The researchers are careful about what the study proves. With a small, single-institution sample, no control group, and remote unsupervised sessions, the findings establish feasibility and acceptability, not pedagogical efficacy. The completeness scorer was even modified partway through data collection, complicating some comparisons. Still, the platform addresses longstanding barriers of cost, logistics, and rigid scripting that have limited traditional standardized patients, and it offers continuous availability with immediate, standardized feedback, qualities especially valuable in resource-constrained settings.

The next step is a controlled trial comparing the system against actor-based training and against a virtual patient without automated feedback, with communication competence measured by blinded human raters before and after the intervention. If the approach holds up, medical students may soon rehearse difficult patients at midnight, receive instant structured critiques, and arrive at their first real ward round having already stumbled through dozens of simulated encounters. The era of practicing medicine on machines before touching patients is no longer hypothetical; it is being scored, one interview at a time.

Subject of Research: LLM-based conversational virtual patients and automated evaluation systems for training medical students in clinical communication and anamnesis

Article Title: Towards clinical communication training with LLM-based conversational virtual patients and automatic evaluation systems

Article References: Bayona, A., Manrique, R., & Grévisse, C. (2026). Towards clinical communication training with LLM-based conversational virtual patients and automatic evaluation systems. Discover Artificial Intelligence, 6(1), Article 1451. https://doi.org/10.1007/s44163-026-02472-w

Image Credits: AI Generated

DOI: 10.1007/s44163-026-02472-w

Keywords: large language models, conversational virtual patient, medical education, clinical communication, anamnesis, automated evaluation, generative agents, semantic memory, LangGraph, usability testing, simulated patients, health communication

The platform at the heart of the work is Artaban, a quadruped robot with twelve degrees of freedom, each leg driven by three Maxon motors. It carries four camera modules, two standard RGB and two depth-sensing RGB-D units, plus a time-of-flight camera in its chest, an inertial measurement unit, motor encoders, temperature sensors, and a built-in speaker. A locomotion controller based on Nonlinear Model Predictive Control translates high-level velocity commands into optimized foot trajectories. Onboard computing comes from an Intel NUC 11 with a Core i7 processor and an NVIDIA RTX 2060 GPU running Ubuntu 20.04 and ROS Noetic.

What makes SMaRTAban distinctive is how it stitches together the pieces of a complete voice-command pipeline. Spoken input passes through voice activity detection adapted from the Google Speech Recognition library, then to OpenAI’s hosted Whisper service, which automatically detects the language and transcribes the audio. The transcript is handed to a ReAct agent built on the LangGraph framework, where GPT-4o reasons step by step and selects from seven custom tools: moving the robot, stopping it, sitting down, standing up, interpreting camera images through GPT-4o vision, synthesizing speech, and waiting. Everything is displayed in a React-based graphical interface so the operator can see every tool call, every response, and every error in real time.

Safety receives unusually careful treatment for a language-driven system. Because large language models are probabilistic by nature, the researchers deliberately avoid relying on the model alone to keep the robot safe. Each tool has a typed schema whose arguments are validated before execution, and invalid parameters are rejected and returned to the model as errors without any physical action. More importantly, nearly every tool runs inside a subprocess guarded by a custom “interruptible” decorator. When the operator presses the interrupt button in the GUI, the main process kills that subprocess and independently issues a zero-velocity command to the robot — no LLM consultation required. Shutting down the application triggers the same stop, and a physical emergency-stop button cuts motor power as a final backstop. The researchers are candid that this layered approach still lacks a controller-level watchdog or velocity timeout, so it reduces risk without providing a formal safety guarantee.

The evaluation is unusually thorough for this field. A pilot voice study collected 120 recordings of four phrases — two short and two long, in both English and Slovak — spoken ten times each by three volunteers. English transcription was highly accurate, with word error rates of just 0.04 to 0.08. Slovak proved much harder at the acoustic level: the short phrase “Choď dopredu” (go forward) reached a word error rate of 0.97, occasionally being misidentified as a different Slavic language. Yet here is the striking part — even when the transcription looked mangled, meaning often survived. The short Slovak phrase still scored 0.87 on BERTScore, a semantic similarity measure, and the system correctly interpreted the intended command in 87 percent of recordings. The longer Slovak phrase reached a BERTScore of 0.96 and 90 percent understanding despite a word error rate of 0.25.

The core of the paper is a controlled ablation spanning 4,800 accepted tool-calling trials: two models, GPT-4o and GPT-4o-mini, across six agent and prompting configurations, two languages, ten command scenarios, and twenty repetitions each. The scenarios covered translation and turning movements, posture changes, a vision query, and one demanding multi-step command requiring the robot to sit, wait, stand, move forward, turn right, stop, and then describe what it sees. The results tell a clear story about what actually makes language-driven robots reliable.

Simply giving the model access to tools was not enough. Without a task-specific system prompt, GPT-4o completed the full expected action sequence in only 29.75 percent of trials. Adding an explicit system prompt describing the action contract — move, then wait, then stop — lifted that to 52.50 percent. But the decisive jump came from few-shot prompting: three worked examples demonstrating the movement pattern. With iterative ReAct execution, the system prompt, and those demonstrations combined, GPT-4o achieved a tool-call F1 of 0.9977 and 96.5 percent exact-sequence success. GPT-4o-mini followed the same trend, reaching 82 percent exact-sequence success at roughly one-twentieth of the model cost, though it failed to produce a single exact sequence for the complex multi-step command.

That gap between individual correctness and complete sequences is one of the paper’s most important lessons. On the complex scenario, GPT-4o-mini earned a respectable F1 of 0.8764 because most of its individual tool calls were right — but the ordering was wrong, typically invoking the camera before the required final stop. GPT-4o completed 26 of 40 complex sequences, including all 20 Slovak repetitions. The authors argue, convincingly, that robotics evaluations of language models must report task-level sequence success, not just per-call accuracy, because a command with a missing or misplaced stop action can be physically dangerous even when it scores well on superficial metrics.

Latency measurements complete the practical picture. On the deployed robot’s own computer, the first tool callback arrived a median of 2.74 seconds after the nominal end of a recorded utterance, rising to 4.10 seconds at the 95th percentile. By contrast, the local ROS stop-service round trip took a median of just 4.28 milliseconds. In other words, cloud transcription and model inference dominate the response path, while local robot communication is nearly instantaneous. This supports a sensible division of labor: the LLM handles interactive, high-level interpretation, while time-critical stabilization and locomotion stay with deterministic local controllers. The study also verified conversation history retention, hazard detection — the system correctly flagged a carpet cutter with an exposed blade as dangerous — graceful error recovery, and reliable mid-execution interruption.

The authors are careful about the limits of their claims. The functional study used scripted commands in dry-run mode, so physical motion quality was not quantified; the voice study involved only three speakers and four phrases; Slovak difficulty at the transcription stage remains an upstream failure mode when meaning is genuinely lost; and dependence on remote APIs introduces network latency and availability concerns. Still, SMaRTAban stands out as the first system in its field to report an evaluated combination of bilingual voice control in English and Slovak, on-demand GPT-4o vision, operator-controlled interruption of running actions, and rigorous quantitative analysis of the whole command path. Future work points toward native multimodal audio input, locally hosted open-source models to remove cloud dependency, the Model Context Protocol for portable tool definitions, integration with SLAM for autonomous navigation, and formal human-robot interaction studies with non-expert users. For now, the image of a quadruped robot that understands “Choď dopredu” as readily as “Go forward” — and stops the instant you tell it to — marks a meaningful step from Čapek’s century-old fiction toward an everyday robotic companion.

Subject of Research: A voice-controlled large language model agent enabling bilingual natural-language control of a quadruped mobile robot with integrated vision and safety mechanisms.

Article Title: SMaRTAban: a voice-controlled LLM agent for quadruped mobile robotics with integrated vision

Article References: Zelenay, E., Kocúr, M., Lukáč, M., Duchoň, F., & Marko, R. (2026). SMaRTAban: a voice-controlled LLM agent for quadruped mobile robotics with integrated vision. International Journal of Intelligent Robotics and Applications. https://doi.org/10.1007/s41315-026-00593-0

Image Credits: AI Generated

DOI: 10.1007/s41315-026-00593-0

Keywords: large language models, quadruped robotics, voice control, Whisper, GPT-4o, ReAct agent, human-robot interaction, natural language interface, LangGraph, robot safety, computer vision, ROS

The motivation stems from a persistent and costly bottleneck. Cross-domain research and development in urban mobility—where batteries, artificial intelligence, vehicle-to-everything communication, and logistics systems must converge—has long been stalled by disciplinary silos. Traditional foresight methods such as expert panels, technology roadmaps, and bibliometric analyses can identify promising trends, but they rarely translate those trends into actionable, structured collaboration plans that pair the right industrial innovators with the right academic partners. Lee’s framework is designed to close that gap by converting patent-based convergence signals directly into industry–academia R&D proposals.

Technically, the system operates in several tightly coupled stages. First, a Bayesian long short-term memory (LSTM) network performs walk-forward signal detection on patent classification data, identifying high-potential technology pairs whose convergence trajectories suggest fertile ground for joint research. The walk-forward design is deliberately conservative: the model is trained only on earlier periods and tested on immediately following held-out periods, with no access to target data during training. In the short-window setting—training on 2000–2014 and predicting 2015–2019—the Bayesian LSTM achieved perfect directional accuracy on six validated technology pairs, outperforming four transparent baseline models under identical conditions.

Once promising technology pairs are identified, the framework builds expert personas directly from inventor records in the patent corpus. Rather than relying on simple patent counts, Lee employs a Linear Weighted Score that recency-weights each inventor’s contributions: a patent filed in 2024 receives roughly 25 times the weight of one filed in 2000. This ensures that simulated experts reflect the current innovation frontier rather than historical output. Sensitivity checks confirmed the robustness of this approach—replacing the weighted score with raw counts or exponential weighting shifted top-10 expert rankings by at most two positions in most cases. The distribution analysis also revealed a striking concentration of expertise: both productivity and influence are heavily right-skewed, with high-impact innovators being exceptionally rare.

The heart of the system is a three-phase dialogue orchestrated through LangGraph, a framework for coordinating multi-agent workflows. In Phase 1, two industry expert agents—constructed from real inventor data, such as specialists in AI-driven smart parking systems and notification control technologies—hold a virtual convergence meeting under one of three facilitation strategies: Consensus-Driven, Greedy-Exploitation, or Exploratory-Brainstorming. The goal is to identify the single most critical academic research field needed to enable a proposed technology fusion. In Phase 2, the system queries ArXiv to discover a suitable academic collaborator, synthesizing the literature into a detailed persona. In Phase 3, the industry and academic agents convene to produce a structured joint R&D plan with quarterly milestones, role divisions, and expected outcomes, output as a parsable JSON meeting log.

The results from 30 capability-demonstration runs were striking. ArXiv retrieval succeeded on the first or second query in all 30 runs, with independent raters scoring persona-to-knowledge-gap alignment at a mean of 4.1 out of 5.0. Every refined proposal incorporated at least one frontier AI paradigm—foundation models, federated learning, quantum machine learning, or agentic intelligence—compared with none at the initial convergence stage, illustrating what Lee terms the candidate synthesis effect. Only one run fell below the quality threshold, and a structural analysis traced the failure to a mismatch between the Greedy-Exploitation strategy and the specific technology pair, producing a proposal with too few milestones to be actionable.

What distinguishes the study methodologically is its Two-Track evaluation architecture, designed to confront one of the most serious criticisms of LLM-based research: the ‘LLM-as-judge’ circularity problem, in which the same family of generative models both produces and evaluates the output. Track A demonstrates the framework’s generative ceiling using GPT-4-class models. Track B, the primary validity evidence, employs six non-generative discriminative scorers—none sharing computational lineage with the generative models—to produce a Collaborative Quality Score. This design substantially reduces evaluator–generator circularity, yielding a pipeline advantage of Δ = +0.275 with a large effect size (d = 2.227) over strong baselines, including a cross-family comparison against Alibaba’s Qwen2.5-3B under both chain-of-thought and direct prompting conditions.

Human validation reinforced the automated findings. Three independent domain experts—an AI/ML professor, a senior ITS/V2X research engineer, and an innovation policy specialist—blind-rated 24 proposals using a 13-item rubric, achieving an intraclass correlation of 0.918, indicating strong inter-rater reliability. An FDR-corrected ablation analysis at 35 runs per condition showed that removing academic integration or LSTM forecasting produced large-effect degradations in proposal quality, while removing personas caused medium-effect degradation—evidence that the pipeline’s modules work synergistically rather than redundantly. Notably, a calibration gap emerged: automated scores were systematically higher than human ratings, and sub-dimension correlations between proxy scorers and holistic human judgments were low, an honest limitation the study documents in detail.

Lee is candid about the framework’s boundaries. Signal detection operates over relatively short windows, automated scoring remains proxy-level, and real-world feasibility of the generated proposals has not yet been confirmed—no simulated collaboration has been executed by actual laboratories or companies. The generalization check comparing centrality-selected versus randomly sampled technology pairs is explicitly flagged as confounded and non-confirmatory, with the author calling for replication with at least 20 pairs per group under a common evaluation model. These caveats, far from undermining the work, reflect a design science ethos in which transparency about limitations accompanies every claim.

The implications nonetheless extend well beyond urban mobility. If patent data can seed credible expert personas, if retrieval systems can locate genuine academic counterparts, and if orchestrated multi-agent dialogue can produce structured, evaluable R&D roadmaps, then the same architecture could be applied to energy transition, biotechnology, semiconductor design, or any domain where convergence across disciplinary boundaries determines the pace of innovation. All datasets and simulation outputs have been released in a public repository, inviting replication and extension. As generative AI matures from a tool for drafting text into infrastructure for planning discovery itself, this study offers a rigorous, carefully validated template for what simulated innovation ecosystems might look like—and a sober reminder that trustworthy evaluation, not generation alone, is what will make them useful.

Subject of Research: A data-grounded multi-agent large language model framework for planning industry–academia R&D collaboration in urban mobility using patent-based technology convergence signals.

Article Title: Simulating the Future of Innovation: A Data-Grounded, Multi-Agent LLM Framework for R&D Collaboration Planning in Urban Mobility

Article References: Lee, Y.-J. (2026). Simulating the Future of Innovation: A Data-Grounded, Multi-Agent LLM Framework for R&D Collaboration Planning in Urban Mobility. Information Systems Frontiers. https://doi.org/10.1007/s10796-026-10797-1

Image Credits: AI Generated

DOI: 10.1007/s10796-026-10797-1

Keywords: urban mobility, multi-agent LLM, R&D collaboration, patent analysis, technology convergence, Bayesian LSTM, innovation simulation, LangGraph, ArXiv retrieval, expert personas, large language models, technology forecasting