Embodied AI for Scientific Discovery: From Physical Action to Scientific Evidence

Wenhao Xu; Changwei Wang; Shunpeng Chen; Yukun Song; Sifan Zhou; Kaize Luo; Shibiao Xu; Yuehua Zhao

cs.AINov 3, 2026

Embodied AI for Scientific Discovery: From Physical Action to Scientific Evidence

Authors: Wenhao Xu, Changwei Wang, Shunpeng Chen, Yukun Song, Sifan Zhou, Kaize Luo, Shibiao Xu, Yuehua Zhao,

Organizations: Zhengzhou Police University, Zhengzhou, China · Qilu University of Technology (Shandong Academy of Sciences) and Shandong Computer Science Center, Jinan, China · School of Artificial Intelligence, Beijing University of Posts and Telecommunications, Beijing, China · Southeast University, Nanjing, China ·

Abstract

Scientific experiments are increasingly selected, executed and interpreted by learned systems that act through physical interfaces. Robot learning builds physical competence, whereas AI for science chooses which experiment to run, each reporting different endpoints. This review treats their interface as one problem, embodied AI for science, where the experiment itself is partially observed: specimen state, apparatus state and measurement processes are latent, actions are costly and irreversible, and missingness depends on execution and outcome. We organize the literature by three closures operating at distinct levels: operational closure (the experiment realized is the one intended), epistemic closure (the result updates beliefs or selects subsequent experiments) and evidential closure (the record supports a claim). We analyze workflows through a record-linked evidence cycle - Record, Diagnose, Calibrate, Confirm - as analytic functions with feedback. A taxonomy crosses these closures with six delegated decisions and five embodiment classes. Our formulation separates commanded, realized and observed conditions and models diagnosis with explicit resource, risk and sample-change costs. We compare experiment-selection methods by terminal decision, review perception, protocol compilation, visuomotor policies, world models, recovery and transfer by evidence emission requirements, and compare chemistry, materials, biology, microscopy and field workflows by what their records establish. An appendix provides an evidence map of 28 workflows (26 physical, two software-boundary), characterizing validation pathways and claim confirmation requirements. We propose evaluation that separates actions, valid experiments, campaigns and confirmed claims, framing open challenges as controlled comparisons with measurable failure conditions.

Figures & tables

Figure 1: Three closures around one shared physical experiment. Each row pairs a closure with its realized state, a three-step loop with an error feedback arc, its characteristic failure and the reporting unit it counts. Operational closure compares commanded with realized conditions and recovers from tolerance violations; epistemic closure updates belief from observations and chooses the next experiment; evidential closure turns trial records into candidate claims that need fresh independent confirmation. All three loops read the shared experiment history at the base. The reporting units

N_{\mathrm{attempted}}

N_{\mathrm{sel}}

and

N_{\mathrm{conf}}

shown here are distinct, and

N_{\mathrm{valid}}

(Section 8 ) is a further unit; no common denominator or one-to-one mapping across trial and claim units is implied. The diagram is conceptual and reports no measured data.

Perspective	Main emphasis	Cross-index used here
Broad AI-for-science reviews [ 24 ]	Representations, prediction and generation	Proposal, physical realization and evidence status.
SDL surveys [ 25 ]	Algorithms, apparatus and chemical practice	Scientific/operational feedback, validity and confirmation.
LLM/agentic surveys [ 26 ]	Process stages, architectures, progress and metrics	Outputs placed within physical execution and evidence boundaries.
Embodied AI [ 27 ]	Computation aligned with the physical world	Capabilities mapped to scientific endpoints and validity.
Focused perspectives [ 28 , 29 ]	Chemical laboratories; data-driven microscopy	Decision–action–interpretation links across domains.
Performance metrics [ 30 ]	Transparent laboratory comparison	Completion, valid data, progress and confirmation distinct.

Table 1: Representative reviews and the analytical connections developed here.

Figure 2: Historical profile of the 48 representative systems by scientific domain and publication period. Each cell gives the number of systems; (1p) and (2p) mark arXiv preprints already included in that count. Row and column totals make the cohort size explicit. Publication status follows the records selected in the accompanying dataset: Co-Scientist uses its published Nature version, whereas

\pi_{0}

uses its arXiv record. The cohort contains 40 journal articles, five conference papers and three arXiv preprints. Periods are ordered categories, not an annual timeline or an estimate of field-wide growth.

Figure 3: Six delegated decisions as complementary scientific interfaces rather than an automation ladder. Each panel pairs a decision with a schematic glyph and the record it must leave; the glyphs are abstract illustrations, not measured data or depictions of particular apparatus. The bottom Shared record bar represents a unified trial ledger across all six decisions; connecting stems are schematic, and the absence of a drawn stem beneath a panel does not imply that decision omits a record. Access and Prepare primarily serve operational closure; Select serves epistemic closure; Acquire spans operational and epistemic closure; Interpret and Confirm serve evidential closure. A workflow may delegate a subset of the six decisions, and the panels are not a sequence.

Decision	Method families (section)	Characteristic failure	Record required for interpretation
Operational closure: intended $\leftrightarrow$ realized
Access	Navigation and mobile manipulation; scheduling and orchestration (Sec. 6 )	Handover changes specimen location or identity without a linked record	Operation identifier; station and handover trace; receiving-station check
Prepare	Visuomotor and force-aware policies, protocol compilation, learned recovery (Sec. 6 )	Command accepted while the realized quantity leaves tolerance	Realized condition with measurement; protocol version; exceptions and lineage
Epistemic closure: belief $\rightarrow$ next experiment
Select	BO, active learning, BED, RL/MPC, language-model planners (Sec. 5 )	Terminal loss fails to represent the scientific question	Objective, feasible set, budget, selection history and every attempt
Acquire	Adaptive sampling, event triggering, signal-channel choice (Secs. 5 , 6 )	Policy-dependent sampling leaves the target domain under-observed	Selection probabilities; acquisition context; independent held-out acquisitions

Table 2: Six delegated decisions grouped by the closure they serve. Each row gives the method families that handle the decision (with section), the characteristic failure when it is mishandled, and the record required before its output can be interpreted.

Figure 4: Causal execution, observed records and specimen lineage. A decision compiles to a command and requested measurement configuration. The command acts through a latent realized condition and an updated specimen–apparatus state before observations are generated; measurement configuration may affect both state and sensing. A dashed inference link indicates that observed evidence constrains, rather than equals, the latent condition. The trial record stores commands, requested configuration, observations, metadata, audit decisions and trial status; a noisy condition estimate can be part of metadata. An interrupted transfer stays in lineage and a remade specimen receives a new identifier. Record state is distinct from claim state. Apparatus imagery is synthetic.

Figure 5: Four terminal decisions determine the epistemic method comparison. Bayesian optimization recommends a design that requires independent verification; active learning supplies a response map assessed on a declared held-out distribution; Bayesian experimental design distinguishes decision-relevant alternatives; reinforcement learning and model-predictive control choose state-dependent action sequences. Their characteristic failures are different, so method choice and terminal loss must be tied to the scientific endpoint. Curves, maps and trajectories are conceptual illustrations with no empirical performance values.

Framework	Terminal decision	Assumption required	Evidence and failure observed/reported	Decision rule: use when / fails when
BO	Best feasible response or Pareto trade-off; regret on a verified recommendation	Response is defined under qualified preparation/measurement; feasible region and observation noise are modeled	Multi-fidelity evaluation finds value dependent on fidelity cost and informativeness [ 78 ] ; noisy maxima and unknown feasibility require independent verification and explicit failure handling [ 20 ]	Use when endpoint and feasible designs are measurable; proxy, validity or feasibility errors change the response functional
Active learning	Predictive or decision-relevant loss over a declared target distribution	Target population is specified independently of acquisition; labels represent that target	Materials mapping, biological optimization and interpretable latent models illustrate coverage and representation learning [ 76 , 83 , 75 ] ; high variance can select irreducible noise or irrelevant regions	Use for prediction or coverage over a declared domain; target or label selection makes the fitted model conditional on the acquired subset
Bayesian experimental design	Choose among explicit hypotheses by expected information or terminal posterior decision loss	Candidate mechanisms are represented and make distinct predictions under feasible interventions/measurements	Robot-scientist and biological model-development examples connect hypotheses to experiments [ 33 , 74 ] ; identical observable distributions remain unresolved [ 32 ]	Use for discrimination among identifiable alternatives; equivalence or omitted mechanisms leave posterior concentration insufficient for claim support
RL / MPC	Minimize control cost across a state-dependent trajectory while preserving the scientific endpoint	State/belief is sufficient for control; reward and dynamics track the endpoint; interaction/reset budget is available	AlphaFlow demonstrates sequential multistep decisions [ 63 ] ; safety guarantees remain conditional on the model [ 19 , 84 ]	Use when intermediate state and action order change later options; reward mismatch, partial observability or model shift can separate return from scientific loss
LLM / agent planners	Propose hypotheses, plans, code or tool calls; a numerical and physical procedure assesses the scientific endpoint	Tool/API access, valid protocol constraints and external statistical/execution checks are supplied	Reported benchmarks score task, code or reproduction outcomes; physical claim assessment requires an independently specified preparation or measurement [ 9 , 10 ]	Use for literature-grounded proposals and workflow coordination; task success does not substitute for uncertainty calibration or claim confirmation

Table 3: Decision frameworks: terminal decision, required assumptions, observed evidence and use rule. Each row links a method family to the assumptions required for its scientific decision and to evidence from cited studies.

Figure 6: Operational stack for a physical workcell. Language protocol is an input, not a stack layer: it emits the intent trace. Five operational layers—perception, protocol compilation, visuomotor policy, embodiment, and verification and recovery—emit, respectively, a state estimate, typed operation, action trace, recorded condition estimate and recovery trace into the shared trial record, from which evidence obligations are assessed. World models, teleoperation and digital twins support these layers and do not replace physical measurement. Sensing modalities include cameras, tactile gel, force–torque sensing and balance telemetry. The apparatus is a synthetic illustrative scene.

Component	Established in cited work	Scientific obligation and checkable evidence	Boundary and the observation that closes it
Perception and state estimation	VLM localization and inspection in labs; cross-sensor touch representations; force as a policy modality [ 72 , 69 , 71 ]	Estimate the protocol-dependent quantity (meniscus, delivered mass, contact, torque); report error against an independent reference under occlusion, reflection and transparent labware; track calibration drift	Pair each percept with an independent sensor reading and report error on the operation’s dependent variable, not pose accuracy
Language planning and protocol compilation	Affordance-grounded plans; executable policy code; laboratory tool-bound planning [ 5 , 6 , 99 ]	Emit a typed protocol with units, identity, preconditions, acceptance criteria, safety authorization and rejected alternatives; link the compiled version to realized effects	Treat compilation as a verified transformation: record unresolved parameters and rejected alternatives, and check acceptance criteria against measurements rather than fluent explanation
Visuomotor policies	Action chunking for fine bimanual tasks; flow action experts; compressed action tokenization [ 1 , 2 , 102 ]	Report delivered quantity, contact tolerance, valid-observation rate, latency, interventions and reset cost, compared with protocol tolerance	Score policies on realized $x_{t}$ and specimen integrity; state the action-representation and latency choices (tokens vs. continuous experts, chunk length) that produced the result
World models and learned simulators	Model-based control; simulator-trained robot policies; joint policy–world-model training [ 3 , 4 , 104 ]	Predict measured scientific variables under held-out interventions with horizon-dependent uncertainty; report paired simulated–physical agreement at the same intervention	Evaluate rollouts on measured quantities rather than pixels, and separate execution, specimen and measurement dynamics; an execution model for $T_{\phi}$ is the near-term fit
Failure detection and recovery	VLM failure detection and explanation; world-model-latent monitors; laboratory closed-loop inspection [ 105 , 106 , 72 ]	Separate execution fault, ambiguous measurement and valid negative; report false-accept and miss rates against declared reference criteria, recovery actions, sample loss and new lineage records	Calibrate detectors against physical references and declare retry semantics per operation: idempotent, compensable or sample-consuming
Transfer and teleoperation	Cross-embodiment policy training; teleoperation; paired sim-to-real evaluation [ 109 , 1 , 52 ]	Preserve intervention and observation semantics across apparatus; report endpoint agreement within a prespecified equivalence margin on a named held-out unit, with adaptation data, labor and failed attempts	Test the scientific endpoint, not the motion, on the held-out apparatus or laboratory, with the margin fixed in advance

Table 4: Operational-layer capabilities, experimental variables and evaluation requirements. Columns distinguish reported component results from the physical measurements needed to evaluate their use in scientific experiments.

Measurement decision	Scientific quantity at risk	Validation required
Select event-containing regions [ 53 ]	Counts, morphology, composition	Propagate selection and boundary errors to the endpoint; test new batches and imaging conditions
Trigger acquisition [ 53 ]	Event frequency, duration, sampled population	Compare independent windows; assess missed or unfamiliar events
Select spatial measurements [ 56 , 76 ]	Response maps, transitions, rare regions	Use held-out acquisitions; account for measurement and motion costs
Choose a signal channel [ 127 ]	Relationships between physical signals	Check calibration and held-out measurements; separate association from mechanism
Denoise or predict image content [ 96 , 97 ]	Weak structures and quantitative signals	Preserve acquired data; verify model-derived content independently
Accept a scientific identity [ 35 , 18 ]	Product or phase assignment	Use complementary characterization when the primary measurement cannot distinguish alternatives

Table 5: Measurement decisions and validation requirements in embodied AI for science

Study and objective	Physical interface	Feedback-driven decision	Demonstrated boundary
Mobile chemist [ 8 ] Photocatalyst search	Mobile manipulator and laboratory stations	Select formulations from measured performance	Researcher-defined objective and search space
Exploratory chemistry [ 35 ] Reaction screening	Mobile robots, synthesis, LC–MS and NMR	Select follow-up reactions and repeat screening hits	Heuristic decisions in specified chemical settings
RoboChem [ 62 ] Photocatalysis	Liquid handling, flow reactor and inline NMR	Bayesian selection of reaction conditions	Stocks and product isolation supplied by researchers
A-Lab [ 17 , 18 ] Inorganic synthesis	Powder handling, furnaces, XRD	Refine recipes from synthesis outcomes	Original count; correction narrows phase and novelty claims
ORGANA [ 36 ] Laboratory assistance	Robots, visual feedback and lab devices	Plan and execute tasks with chemist interaction	Assistive workflow; human disambiguation remains explicit
ChemAgents [ 92 ] Multiagent chemical research	Automated lab, libraries, robot operator	Coordinate literature, design, computation and execution	Six tasks plus one in another lab; independent confirmation unestablished

Table 6: Chemistry, materials and energy: physical integration and scientific decision scope

Study and objective	Physical interface	Feedback-driven decision	Demonstrated boundary
Robot scientists [ 33 , 7 ] Gene-function hypotheses	Laboratory robotics and growth assays	Select experiments and update explicit biological hypotheses	Bounded by supplied models and available assays
BioAutomata [ 124 ] Biosystems design	Integrated biological foundry	Iterative design–build–test–learn selection	Pathway optimization rather than general mechanistic discovery
Cell-culture search [ 125 ]	Robotic cell handling and assessment	Search culture conditions from experimental feedback	Biological state and preparation history remain relevant
SAMPLE [ 57 ] Protein engineering	Gene assembly, cell-free expression and assays	Update sequence–function models and select proteins	Supplied library and assay; follow-up in an alternative context
METIS [ 83 ] Network optimization	Manual experiments or liquid-handler integration	Iterative suggestions from measured biological objectives	Selection workflow; physical integration depends on the application
Event-driven imaging [ 53 ]	Programmable fluorescence microscope	Change acquisition rate when an event is detected	Adaptive acquisition; detection can shape the sampled population

Table 7: Biology, adaptive measurement and field science: scientific and operational boundaries

Dataset or benchmark (examples)	Contribution	What it measures	Scientific question left open
Robot training datasets: Open X-Embodiment and DROID [ 110 , 111 ]	Demonstration data for policy and representation training	Training coverage, not scientific fidelity	Whether a policy action produces a valid intervention or measurement
Robot benchmarks: LIBERO and SIMPLER [ 51 , 52 ]	Task suites for knowledge transfer and paired sim–real policy evaluation	Task success and transfer	Whether success implies quantitative delivery, containment, sterility or sample integrity
Policy and world-model evaluation: RoboArena and WorldArena [ 129 , 13 ]	Distributed pairwise real-world policy comparison; perceptual and functional evaluation of world models	Relative policy ranking and proxy agreement with real runs	Whether the ranking holds for a measured physical endpoint rather than a task label
Language-agent benchmark for biology: LAB-Bench [ 130 ]	Multiple-choice questions on literature, figures, databases and sequences	Practical research knowledge and reasoning	Whether a correct answer yields a correct physical intervention
Optimizer benchmarks: Olympus and its extensions [ 80 , 81 ]	Controlled problems for comparing selectors	Selector behavior	Whether a selected experiment is reachable, feasible and informative for the target quantity
Agent benchmarks: ScienceAgentBench and CORE-Bench [ 9 , 10 ]	Scientific data-analysis and computational-reproduction tasks	Task, code or reproduction outcomes	Whether the artifact is true of, or useful for, a physical system

Table 8: Datasets and benchmarks for embodied scientific systems, grouped by the question they answer

Unit	Report	Why this measure is needed
Action	Tolerance error; completion; collision/containment events	Distinguishes meeting an operation specification from obtaining a valid observation.
Experiment	$N_{\mathrm{valid}}/N_{\mathrm{attempted}}$ ; missing/inconclusive outcomes; raw-data availability	Retaining all started attempts prevents failures disappearing from the reliability denominator.
Measurement	Calibration; noise; detection limits; reference samples	Separates instrument repeatability and sensitivity from between-preparation variation.
Recovery	Interventions; recovery time; remade samples	Reveals whether recovery preserved the specimen and how much human work completion required.
Campaign	Endpoint versus attempts, time and resources	Prevents experiment count or operating speed from standing in for endpoint progress.
Transfer	Adaptation data/labor; held-out endpoint agreement	Charges the information and labor needed to move beyond the training context.

Table 9: Reporting quantities matched to the unit of scientific work

Metric	Definition	Inputs	Truth requirement and interpretation
Valid-observation rate	$\sum_{t}\mathbb{1}\{\hat{v}_{t}=1\}/N_{\mathrm{attempted}}$	Audit-adjudicated trial records	Online operational audit indicator; no privileged truth required.
ECYG	Eq. 18	$U_{T}^{\mathrm{claim}}$ , $\sigma_{k}$ , declared budget $B$	Confirmed scientific yield per total budget; requires a prespecified confirmation rule.
COR	Eq. 19	Claims presented as confirmed and their $\sigma_{k}$	Post-hoc claim overstatement rate; unresolved claims remain non-confirmed.
CSU	Eq. 20	Claim utility weights $w_{k}$ , $\sigma_{k}$ , $B$	Utility-weighted confirmed yield; weights are fixed before comparison.
DA	Eq. 21	$\hat{v}_{t}$ , protocol-relative $v_{t}^{\star}$	Controlled injection benchmark only; agreement with the protocol’s reference criteria.
IRR	Eq. 22	$\hat{v}_{t}$ , protocol-relative $v_{t}^{\star}$	Controlled injection benchmark only; rejection of records violating the protocol’s reference criteria.

Table 10: Metrics aligned with trial validity, claim state and scientific utility. Privileged truth labels are used only in controlled injection benchmarks.

Figure 7: Evaluation design for embodied scientific campaigns. A strict two-by-two comparison crosses a reference or adaptive selector with a baseline or improved executor under matched endpoint, information, resource boundary and acceptance criteria. Every attempt enters the physical execution gate and remains in $N_{\mathrm{attempted}}$ . The metrological and audit gate assigns $\hat{v}_{t}$ and $c_{t}$ : valid positive, valid negative and censored records all count in $N_{\mathrm{valid}}$ , whereas execution failures and inconclusive records remain in $N_{\mathrm{attempted}}$ but not $N_{\mathrm{valid}}$ . Nominated claims form a separate unit with counts $N_{\mathrm{sel}}$ and $N_{\mathrm{conf}}$ ; confirmation uses fresh evidence not used for nomination under prespecified criteria. $r_{\mathrm{valid}}$ , ECYG, COR and CSU use their declared denominators; DA and IRR require a privileged reference label and apply only to controlled injection benchmarks. The apparatus is a synthetic illustrative scene.

Figure 8: Failure attribution and claim reassessment. A low response with qualified preparation and adequate detection can be a valid negative; valid positive, valid negative and censored records remain in the valid pool and count in $N_{\mathrm{valid}}$ . Uncertain transfer produces an execution failure that remains in $N_{\mathrm{attempted}}$ and returns through diagnosis and recovery, while an unclear signal is inconclusive. These trial statuses $c_{t}$ are distinct from claim state $\sigma_{k}$ . Independent confirmation of a nominated claim can retain, revise or narrow it according to prespecified evidence; narrowing is not the only permissible outcome. The apparatus is a synthetic illustrative scene.

Item	Minimum content	Why it is needed: failure prevented
Scientific question and endpoint	Target quantity, alternatives/decision and success criterion	Prevents task completion or a proxy score being substituted for the scientific endpoint.
Initial information	Prior data/models, constraints, search space and system inputs	Prevents unequal information or training exposure being credited as a method gain.
Action space and preconditions	Controls, allowed operations, material/equipment prerequisites	Distinguishes selector quality from an apparatus that cannot reach the required experiment.
Selector and version	Decision rule, model, code version and training data	Prevents an untraceable policy change from being mistaken for an executor effect or reproducible result.
Executor and apparatus	Configuration, calibration, tolerances and operating envelope	Prevents hardware reliability and calibration differences from being attributed to selection.
Realized conditions	Measured intended quantities, deviations and specimen lineage	Prevents nominal commands from standing in for the actual physical intervention or specimen.

Table 11: Minimum reporting items for embodied scientific campaigns

Item	Mechanism and independent intervention	Controlled comparison or compliance test	Dependent metric	Falsification or violation condition
P1. History-aware state	Add specimen lineage, realized quantities and execution telemetry to the policy state; hold robot, selector and model capacity fixed.	Images alone versus images plus telemetry versus lineage/history, with complete batches held out.	Valid-observation rate; realized-quantity error; held-out-batch endpoint transfer.	A sufficiently precise comparison excludes the prespecified useful gain on tasks where history changes sample state.
P2. Recovery preserves validity	Replace scripted recovery with a correction-trained policy under matched injected faults; hold executor and starting conditions fixed.	Scripted/replay recovery versus learned recovery with the same fault set and resource budget.	Valid recovered experiments; interruption time; sample loss; human burden; ECYG.	Similar recovery with greater burden, damage, invalid records or lower confirmed yield.
P3. Transfer preserves meaning	Adapt one policy or protocol to a declared hardware, apparatus or laboratory change while holding the scientific endpoint and equivalence margin explicit.	Two nonidentical platforms with matched preparations, calibration and adaptation accounting.	Endpoint agreement; adaptation data, labor and calibration cost; COR.	Endpoint disagreement beyond the prespecified margin after adaptation is charged.
P4. Validated capability and co-design expand reach	Add one qualified intervention, measurement or AI-designed apparatus configuration with stated preconditions, uncertainty and failure modes.	Existing vocabulary/fixed apparatus versus the added capability or co-designed apparatus at the same endpoint and total budget.	New question reached; physical reachability; valid endpoint; transfer; total cost.	No new reachable question or valid endpoint at comparable cost, or simulated benefit fails after fabrication and calibration.
P5. Action-aware models improve prediction	Condition the scientific model on the realized intervention and measurement configuration rather than only a nominal command or visual rollout.	Intended-condition, realized-condition and measurement-aware models with matched data and capacity; hold out interventions.	Physical response error; alternative discrimination; uncertainty calibration; horizon.	No gain after instrument effects and realized actions are controlled.
P6. Budget-aware selection improves evidence	Compare matched-budget policies that allocate resources to primary exploration, adaptive diagnosis, orthogonal confirmation and proposal computation.	First isolate a two-way primary/confirmation allocation with integer trial-equivalent budget $N$ ; then compare objective-only selection, adaptive diagnosis, and confirmation-governed selection with all diagnosis and computation charged to the same total resource vector.	Actual ECYG and COR; modeled ICY opportunity-cost benchmark; unresolved outcomes; infeasible-proposal fraction; confirmation, diagnosis and computation cost.	No improvement in the prespecified confirmed-evidence/invalid-report trade-off under matched total cost, or confirmation harms a specimen or biases the blinded assessment.

Table 12: Research priorities as six falsifiable empirical propositions and three normative architectural and reporting invariants. Dependent metrics use the definitions in Tables 10 and 9 .

Figure 9: Six empirical propositions grouped into three complementary research thrusts, with three cross-cutting normative invariants. P1 and P5 concern histories and realized actions; P2 and P6 concern validity-preserving recovery and matched-budget selection, diagnosis and confirmation; P3 and P4 concern endpoint-preserving transfer and qualified expansion of experimental reach. Provenance completeness, safety with human authority, and cross-site replay compliance must hold across all propositions. Grouping does not imply a temporal or causal order, predicted gains or established findings. Scientific miniatures are conceptual and apparatus imagery is synthetic.

AI for Science

AI for Science applies machine learning and AI agents to scientific data, models, experiments, and discovery workflows.

See all 166 papers

Mar 20, 2026

Embodied Science: Closing the Discovery Loop with Agentic Embodied AI

Artificial intelligence has demonstrated remarkable capability in predicting scientific properties, yet scientific discovery remains an inherently physical, long-horizon pursuit governed by experimental cycles. Most current computational approaches are misaligned with this reality, framing discovery as isolated, task-specific predictions rather than continuous interaction with the physical world. Here, we argue for embodied science, a paradigm that reframes scientific discovery as a closed loop tightly coupling agentic reasoning with physical execution. We propose a unified Perception-Language-Action-Discovery (PLAD) framework, wherein embodied agents perceive experimental environments, reason over scientific knowledge, execute physical interventions, and internalize outcomes to drive subsequent exploration. By grounding computational reasoning in robust physical feedback, this approach bridges the gap between digital prediction and empirical validation, offering a roadmap for autonomous discovery systems in the life and chemical sciences.

Xiang Zhuang, Chenyi Zhou, Kehua Feng +10

Zhejiang University, Hangzhou, China · Shanghai Artificial Intelligence Laboratory, Shanghai, China · Tongji University, Shanghai, China

AI Agents for Scientific Discovery Autonomous Experimentation Embodied AI Scientific Reasoning in Language Models

Ask PDF

Aug 13, 2026

OmniScientist: An Omni-Modal Omni-Discipline AI Scientist

Recent advances in foundation models have enabled AI scientists to automate increasingly complete research workflows, from hypothesis generation and code execution to manuscript preparation. Yet workflow coverage alone does not provide access to the full evidence on which scientific discovery depends. Existing systems typically reason over text, code, labels, or precomputed summaries, leaving scientifically decisive spatial, temporal, cross-channel, and procedural relations unavailable to the agent. We introduce OmniScientist, an end-to-end, omni-modal AI scientist that conducts multidisciplinary research directly from heterogeneous raw evidence. A perception layer and 3 autonomous agents for ideation, experiment, and writeup operate within a deterministic pipeline, allowing observations to shape research questions, experimental decisions, and final claims throughout the research lifecycle. By running idea, rigour, and claim checks in code, the system enforces novelty screening, statistical validity, execution provenance, and numerical traceability. We evaluate OmniScientist on 36 real-data cases spanning 5 discipline families, 4 families of scientific evidence, and modalities including images, signals, audio, video, 3-D structures, trajectories, tables, formulae, and graphs. The system completes the full path from raw data to a compiled manuscript in all 36 cases and achieves a mean overall paper score of 6.3 with the reference reasoning backbone. In paired comparisons against a blind variant that receives only precomputed scalar features, direct perception improves all 7 evaluation dimensions and wins 85% of head-to-head judgments. These results show that lifecycle-wide perception is essential for evidence-grounded scientific discovery and provides a practical path toward broadly capable AI scientists.

Bobo Li, Hao Fei, Tianjie Ju +2

1National University of Singapore · University of Oxford

AI Agents for Scientific Discovery Scientific Discovery Autonomous Scientific Discovery Evidence-Grounded Generation Scientific Workflow Automation

Ask PDF

May 6, 2026

Experiment-as-Code Labs: A Declarative Stack for AI-Driven Scientific Discovery

To unleash the full potential of AI for Science, we must untether the agents from a purely digital environment. The agent's ability to control and explore in real-world labs is essential because the physical lab remains foundational to scientific discovery. While some tasks can be performed on a computer (e.g., data analysis, running simulated experiments), Eureka moments could occur at any time while operating lab instruments (e.g., when a scientist notices unexpected clues, intuition may prompt a real-time course change). Although autonomous labs are on the rise, which expose programmable APIs to control scientific instruments via software, bridging the gap between increasingly powerful AI agents and automated lab equipment requires innovation that draws insights from computer systems. We propose a new paradigm called ``Experiment-as-Code (EaC) Labs,'' where a core concept is to encode experiments as declarative configurations that can be compiled down to device-level APIs. AI agents come up with hypotheses and experiments, written as an ensemble of declarative configurations. The systems layer performs program analysis, safety checks, resource assignment, and job orchestration. Finally, programmatic experimentation occurs via actuating the device APIs. This is a general stack that is science-, lab-, and instrument-independent, representing a novel synthesis across the physical, systems, and intelligence layers to unleash the next breakthrough in AI for Science.

Zhenning Yang, Yuhan Chen, Patrick Tser Jern Kon +5

University of Michigan

AI for Science AI Agents for Scientific Discovery Laboratory Automation Scientific Workflow Automation Autonomous Experimentation

Ask PDF

Abstract

AI for Science

Explore similar work

Embodied Science: Closing the Discovery Loop with Agentic Embodied AI

OmniScientist: An Omni-Modal Omni-Discipline AI Scientist

Experiment-as-Code Labs: A Declarative Stack for AI-Driven Scientific Discovery