Together, under a single agent, these components aim to accelerate the SDL loop by reducing both the number of loops and the cost per experiment.
Abstract
Agentic AI-for-Science can automate ideation, planning, and analysis, but final validation still depends on real experiments. A self-driving lab (SDL) can execute those experiments, yet the loop still has bottlenecks: the agent may spend too many rounds on low-value experiments, or each round may require a high-cost experiment. We target these two physical bottlenecks with one agent. First, a prior-aware agentic DOE loop uses domain knowledge and past results to propose feasible and informative next experiments, reducing trials-to-target. Second, a cost-aware surrogate agent predicts high-cost, high-resolution measurements from low-cost, low-resolution measurements. It chooses between a high- and a low-cost measurement based on the predicted uncertainty. We examine these directions in the biology and materials domains, respectively. Together, under a single agent, these components aim to accelerate the SDL loop by reducing both the number of loops and the cost per experiment.
Automating data science on large-scale projects has become a problem of engineering loops rather than tuning models by hand. This paper takes a design stance on that problem and presents LEAP, a loop-engineered agentic pipeline in which an orchestrator agent coordinates stage agents through three feedback loops: a fast loop that searches model configurations under a compute budget, a medium loop in which a critic verifies candidates and revises the pipeline, and a slow loop that monitors and retrains under drift. We make two parts of the design precise. We show that a blind search needs on the order of 1/p trials to find a configuration that is within tolerance of optimal, which motivates cheap multi-fidelity screening, and we show that the optimism of selecting the best of N configurations by validation grows like √2 ln N, which motivates a cross-validated critic. We validate the design with a fully reproducible experiment written from scratch, and we report honest results. At equal compute, agentic multi-fidelity search reaches a test AUC of 0.749 against random search’s 0.738 and wins in 79 percent of runs; a critic loop improves test AUC at every search intensity and curbs the validation optimism that heavy search injects, which rises to 0.037 under the heaviest search; and the critic advantage holds as dimensionality grows. Crucially, a naive greedy agent does not beat random search, which is a strong baseline. The value of agentic automation therefore lies in how it allocates compute and how it selects, not in the promise of a cleverer model, and both can be engineered and reasoned about.
Arunkumar Medisetty· International Journal of Com...· 0 citations
Despite strong capabilities in data understanding and decision-making, autonomous data science agents still heavily rely on trial-and-error workflows that involve expensive computation. This bottleneck motivates models that can anticipate the effects of data science operations before real execution. In this paper, we introduce the concept of Data Science World Model, which model the data science execution environment by predicting environment state transitions conditioned on current workflow states and candidate operations. We further propose DSWorld, a practical framework that combines structured state construction, cost-aware routing, lightweight real execution, and an LLM-based simulator for expensive operations. To support training, we construct an 8K-scale transition trajectory dataset and introduce Reflective World Model Optimization, an error-aware reinforcement learning strategy for improving transition prediction. Experiments show that DSWorld accelerates RL-based agent training by approximately $14\times$ and search-based inference by approximately $3$-$6\times$ while maintaining competitive performance, and outperforms the strongest LLM baseline by 35.6% on transition prediction tasks. The code is available at https://anonymous.4open.science/r/DSWorld.
LabRobFail, a failure-centric framework for learning and evaluating robotic failure analysis in chemical laboratories, and LabRobFail-VLM, a domain-specialized vision-language model that generates structured failure diagnoses and recovery instructions, demonstrate the value of fine-grained failure understanding for closed-loop recovery and reliable laboratory autonomy.
Haobo Wang, Baoli Sun, Anqi Zou et al.· 0 citations
Multi-agent systems in the real-world (e.g., drone swarms, autonomous cars, warehouse robots) must satisfy rich, temporal tasks while avoiding collisions. Signal Temporal Logic (STL) elegantly encodes such objectives, but current STL planning methods face critical limitations. State-of-the-art optimization-based approaches can handle arbitrary STL specifications but struggle with scalability, becoming computationally impractical as the number of agents grows. Learning-based methods efficiently handle a large number of agents with rapid planning times but fare poorly when deployment-time objectives differ from those used during training, and do not support planning tasks that require different specifications to be ascribed to different agents (i.e., heterogeneity) or team-level specifications requiring coordination of multiple agents. This fundamental trade-off between generalizability and scalability presents a challenge for realizing multi-agent STL planning algorithms in practice. To overcome this challenge, we introduce a new diffusion method for multi-agent planning with STL specifications. Using a differentiable approximation of STL, we integrate the STL gradient in the denoising process, making our approach generalizable to novel formulas whose predicates are placed anywhere within the goal region covered during training, while achieving the same scalability as existing learning-based methods. Our method supports heterogeneous specifications, and by using diffusion models, naturally enhances plan diversity, thereby significantly reducing safety-related violations (e.g., collisions) among agents. A detailed evaluation study justifies the utility of STL-guided diffusion-based multi-agent planners for constructing generalizable, scalable, and diverse plans. Videos and code are available at https://www.jeappen.com/diff-ma-stl/ and https://github.com/jeappen/diff-ma-stl .
Joe Eappen, Zikang Xiong, S. Iyengar et al.· IEEE Robotics and Automation...· 0 citations
Although autonomous, Large Language Model (LLM)-driven systems show immense potential for orchestrating complex scientific experiments, their efficacy is constrained by two fundamental bottlenecks: context dilution, where strategic reasoning degrades as experimental history accumulates, and inter-campaign amnesia, which forces systems into computationally expensive tabula rasa explorations upon encountering novel domains. To overcome these limitations, we introduce the Multi-Objective State-Action Network (MOSAN), an autonomous cognitive information fusion framework designed for robust cross-campaign generalization and sensor shift adaptation. MOSAN achieves effective information fusion by coupling a self-evolving semantic memory with an Organic Strategic Graph Memory (OSGM), strictly orchestrated through a novel Single Ledger Architecture (SiLA) that isolates cognitive phases and prevents context saturation. A core mechanism of the OSGM is Cross-Domain Strategic Seeding ("ghost node" injection), a transient cold-start fusion strategy. Upon encountering novel datasets, the OSGM temporarily injects topological priors from similar past domains; these ephemeral nodes guide initial architectural deductions and are subsequently purged, allowing the system to disseminate structural knowledge across disciplines while strictly guaranteeing zero data contamination. The fusion framework was validated on the challenging task of bacterial classification via Surface-Enhanced Raman Spectroscopy (SERS) subject to sensor aging shifts, autonomously discovering highly accurate multi-topological architectures. Crucially, when subjected to an out-of-distribution 1D sequential modality (the ECG5000 cardiovascular dataset), the agent performed zero-shot architectural meta-learning, bypassing brute-force search to rapidly achieve 0.9930 accuracy. Finally, the framework was successfully deployed using an open-weight model (Mistral 24B) at zero API inference cost. By continuously fusing multi-epoch empirical evidence, bridging representational topologies, and actively overcoming sensor drift, MOSAN establishes a scalable and rigorous paradigm for autonomous biophysical discovery.
D. Sagnelli, B. Guilcapi, A. Milano et al.· Machine Learning: Science an...· 0 citations
This paper proposes World Model RL (WMRL), which replaces environment execution with a world model to remove this bottleneck and accelerates training by 3-4x on various tasks at different agent scales, while exceeding the performance of standard RL baselines.
Xi-Yuan Yang, S. Sarwar, Jingru Cheng et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
