{"id":"c8e59fd2-be86-468a-a00d-57b2528eb824","arxiv_id":"2411.18587","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"In a 15-person EEG study, relative alpha power was lower and the theta-alpha ratio higher during a multimodal image-word-motor robot task than during a motor-only task, interpreted as more sustained engagement.","lead":"Researchers compared EEG brain activity in 15 healthy adults during a traditional robot-arm movement task versus a multi-modal task that added image and spoken-word matching. They found lower relative alpha power and higher theta-alpha ratio in the multi-modal task, which they interpret as stronger user engagement, relevant to keeping patients engaged during robotic rehabilitation.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The engagement conclusion rests on an untested mapping from relative alpha suppression to engagement; since relative power is normalized by total 1-100 Hz power, the headline finding could be a normalization artifact, and no behavioral or subjective engagement measure is provided.","rationale":"The reader's weakest-assumption analysis identifies exactly the load-bearing vulnerability: the study validates neither the EEG-to-engagement mapping nor the relative-power normalization. My stress-test confirms this is the most serious point. The statistical multiplicity issue is real but secondary: even a single significant effect would still require the engagement interpretation. The absence of any behavioral or subjective engagement measure means the construct validity rests entirely on external literature, which is a legitimate basis for a pilot but not sufficient for the strong claim that the multimodal task 'modulates engagement.' The proposed concrete test is decisive because it distinguishes between a true alpha suppression and a denominator-driven artifact using data the authors already possess. Since the reader's CONDITIONAL verdict already reflects this concern and asks for exactly this kind of reanalysis, no adjustment to the verdict is needed. I am not raising a new objection; I am endorsing the reader's assessment and making the decisive check explicit.","tokens_in":11459,"tokens_out":1884,"duration_ms":22795,"concrete_test":"Recompute the planning-phase comparisons in Figures 3 and 6 using absolute alpha power (sum of PSD in 8-13 Hz, without dividing by total 1-100 Hz power) for each subject, region, and task, and also test whether total 1-100 Hz power differs between tasks. If the absolute alpha decrease is not significant, or if total power increases substantially in the Matching task, then the relative-alpha effect is at least partly a normalization artifact and the engagement claim is unsupported. As a secondary check, correlate per-subject TAR changes with a basic behavioral engagement measure (e.g., reaction time or accuracy), which would directly test the construct-validity premise.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that lower relative alpha power and higher theta-alpha ratio in the Matching task demonstrate increased engagement. That inference requires two linked premises: (1) alpha suppression is a valid, specific marker of engagement, and (2) the observed decrease in relative alpha reflects actual alpha suppression rather than a change in the denominator. Premise (2) is directly threatened by the paper's own definition in Section II-B-3: relative power is computed by dividing band power by total 1-100 Hz power. If the Matching task produces a broadband increase in EEG power (from visual/auditory processing, movement preparation, or arousal), relative alpha can drop mechanically even when absolute alpha is unchanged or increased. The theta-alpha ratio is less exposed to total-power normalization, but it still depends on the same assumption that theta/alpha changes specifically index engagement. No within-study validation is offered: there are no subjective engagement ratings, no reported task accuracy or reaction times, no manipulation check, and no absolute-power analysis. The paper cites external literature for the alpha-engagement link, but that literature does not establish that the specific task contrast in this study maps onto engagement rather than general cognitive or sensory load. Thus, the load-bearing link between the EEG statistics and the psychological construct of engagement is assumed, not demonstrated, and the relative-power normalization provides a concrete mechanism by which the headline result could arise without any true engagement difference.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents an EEG study with 15 healthy participants who performed a multi-modal 'Matching' robot task (visual image plus spoken word plus motor response) and a motor-only task using the H-Man haptic robot. The authors compute relative power in the delta, theta, and alpha bands and the theta-alpha ratio (TAR) during a 'planning' phase (1 s before movement onset, identified from sEMG) and a 'movement' phase (1 s after onset), then compare the two tasks across six brain regions. They report significantly lower relative alpha power and higher TAR in the Matching task during the planning phase, and a decline in TAR over the session in the motor-only task but not the Matching task. The central claim is that the multi-modal task increases user engagement, as indexed by these EEG biomarkers.","tokens_in":11660,"tokens_out":4021,"duration_ms":37040,"significance":"If the assumed EEG-to-engagement mapping is valid, the result is a useful proof-of-concept for designing robotic rehabilitation protocols that sustain engagement; the within-subject design, the 100-trial-per-condition protocol, the sEMG-based movement-onset detection, and the multi-region analysis are notable strengths. The subject-level consistency (93-100% of participants following the reported trends for the main planning-phase effects) is also a positive feature. However, the central inference depends on an external mapping from relative alpha suppression and TAR increases to engagement, and the relative-power normalization creates a specific artifact risk that the paper does not address. The significance is therefore conditional on validation (absolute-power analysis or a behavioral/subjective engagement measure) that the manuscript does not currently provide.","major_comments":[{"comment":"The definition of relative power in Section II-B-3 (band PSD summed and divided by total 1-100 Hz PSD) means that a broadband increase in total power during the Matching task, for example from visual and speech processing or general arousal, would mechanically lower relative alpha power even if absolute alpha power were unchanged or increased. The paper reports only relative power and never reports absolute band power, so the statement in Section III-A that the data 'strongly support the hypothesis that the users are more engaged' is not directly supported by the presented quantities. Please report absolute band power or otherwise demonstrate that the effect is specific to alpha suppression rather than to normalization by total power.","section":"Section II-B-3, Figs. 2-5"},{"comment":"The interpretation of every result as 'engagement' rests on the assumption, taken from refs. [26]-[31], that decreased alpha and increased theta/delta/TAR index engagement. The manuscript provides no within-study validation of this mapping: there is no subjective engagement rating, no behavioral performance measure (accuracy or reaction time), and no manipulation check. Because the Matching task differs from the motor-only task in visual, auditory, and cognitive demands, the observed EEG differences may reflect generic sensory or cognitive load rather than engagement per se. Please add at least one validation measure, or substantially temper the conclusion to state that the multi-modal task modulates EEG markers 'associated with' engagement, leaving construct validation as future work.","section":"Introduction, Section III-A, Section IV"},{"comment":"The analysis involves a large number of statistical tests (three frequency bands, five to six brain regions plus whole brain, two phases, and the over-time TAR comparison), but no correction for multiple comparisons is applied; instead, four nominal significance thresholds are reported. Given this multiplicity, the pattern of significant results, particularly in the movement phase, should be interpreted with caution. Please report corrected p-values (e.g., FDR) or explicitly frame the analysis as exploratory before drawing the strong conclusion in Section IV that the biomarkers 'showed increased engagement during the Matching task.'","section":"Section II-B-4 and Section III"}],"minor_comments":[{"comment":"There is a typo: 'The Motor-Only results are shown on the right, and the Matching results are on the right' should likely read 'on the left' and 'on the right'; please correct the figure description.","section":"Section III-A, first paragraph"},{"comment":"The damping coefficient is given as 125 N/ms; the SI unit for damping is typically N·s/m (or N·s/m), so please check and correct the unit.","section":"Section II-A"},{"comment":"The phrase 'statistically significant improvements in engagement' conflates statistical significance with a value judgment; consider rephrasing to 'statistically significant increases in EEG engagement markers.'","section":"Abstract and Section IV"},{"comment":"The text says 'the EEGLAB run ica function was used to identify and remove artifacts'; more precisely, ICA identifies components and then artifact components are selected for removal, so the wording could be clarified.","section":"Section II-B-1"},{"comment":"The paper would benefit from reporting effect sizes or confidence intervals for the key planning-phase comparisons (e.g., mean relative alpha at POz), since the significance thresholds alone do not convey the magnitude of the effects.","section":"General"}],"recommendation":"major_revision","confidential_remarks":"The paper is not circular in the sense of fitting parameters to data, but the central claim is construct-validity-limited: the EEG-engagement mapping is assumed from the literature and the relative-power normalization adds a specific artifact risk. The requested changes (absolute-power analysis, validation or reframing, and multiple-comparison awareness) are within the scope of a revision and should determine whether the claim can be supported."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a clean within-subject EEG comparison of a multi-modal robot task against a motor-only control, and the relative alpha and theta-alpha differences are large and consistent across participants. The genuinely new part is the specific combination—visual, word, movement—and the first EEG look at that package. But the paper labels the effect 'engagement' without measuring engagement. That mapping is borrowed from prior literature, and the relative-power normalization gives a concrete route by which the headline effect could appear even if absolute alpha did not change. It is a solid pilot, and it deserves a serious referee with required revisions.\n\nWhat the paper does well: the methods are unusually transparent. sEMG-based epoching is a nice touch, separating planning from movement phases is informative, and they report per-participant consistency (93–100% following the trend) rather than hiding behind group means. The theta-alpha ratio is less exposed to the total-power normalization artifact than the relative alpha result, and the planning-phase differences are big. For a 15-subject pilot, this is careful work.\n\nSoft spots, in rough order of importance. First, no within-study validation of the engagement construct: no subjective rating, no task accuracy or reaction time, no absolute power analysis. Without those, the EEG differences can be read as cognitive load, sensory processing, or task difficulty. The paper does not mention this limitation. Second, the relative power denominator is total 1–100 Hz power; a broadband increase in the matching task could lower relative alpha mechanically. They should report absolute band power, especially for alpha where the claim is strongest. Third, they run many region–band–phase comparisons at multiple alpha levels without correction. The consistency of the effects helps, but an FDR or permutation control is needed. Fourth, the 'maintains engagement over time' conclusion leans on a null result in the matching task (no significant TAR decline), but there is no interaction test, and the motor-only planning-phase decline is not significant either. That claim is stronger than the data. Fifth, the delta band interpretation is the weakest: delta is artifact-prone and its engagement link is less established.\n\nWho this is for: researchers in rehabilitation robotics, EEG-based engagement metrics, and adaptive human-robot interaction. It is a useful pilot and a good cautionary example about relative power. I would send it to peer review, with a strong request for absolute power, a behavioral or subjective engagement anchor, corrected statistics, and more temperate language about what 'engagement' means. If the authors deliver that, this becomes a solid contribution.","headline":"Clean within-subject EEG contrast between multimodal and motor-only tasks, but the engagement interpretation rests on an untested mapping and relative-power normalization; send to review with required revisions.","tokens_in":12208,"tokens_out":5018,"would_cite":true,"duration_ms":47391,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A robotic task that combines image, word, and movement demands produces measurably higher EEG engagement markers than a motor-only task, with the effect strongest during the planning phase.","keywords":["EEG","engagement biomarkers","human-robot interaction","robotic rehabilitation","relative alpha power","theta-alpha ratio","multi-modal task","motor learning"],"falsifier":"A decisive test would run the same protocol while measuring absolute (non-normalized) alpha power plus a subjective engagement rating or behavioral performance metric; if absolute alpha does not differ between tasks, or if perceived engagement fails to track the EEG differences, the central claim would not be supported.","tokens_in":11238,"feed_emoji":"🧠","tokens_out":5277,"duration_ms":45593,"temperature":0.7,"pith_summary":"This paper argues that combining visual, auditory, cognitive, and motor demands in a single robot-driven task engages the brain more than a motor-only task, and that the effect shows up in objective EEG markers. Fifteen healthy adults performed 100 matching trials, where they moved a robot handle to indicate whether a spoken word matched an on-screen image, and 100 motor-only trials, where they moved the handle to a target. In the one-second planning window before movement, the matching task produced significantly lower relative alpha power across the brain and a significantly higher theta-alpha ratio in every brain region, with the majority of participants following the trend. The matching task also kept its theta-alpha ratio stable across the session, while the motor-only task showed a decline. The authors present this as the first neural evidence that a comprehensive multi-modal robotic intervention increases engagement in healthy subjects, which matters because engagement is thought to drive neuroplasticity and adherence in rehabilitation.","feed_headline":"Multimodal robot tasks lift EEG engagement over motor-only drills","feed_subtitle":"Lower relative alpha and stable theta-alpha ratios in healthy adults point to longer, more engaged robot therapy sessions.","key_machinery":"The argument runs on frequency-band power ratios computed from 64-channel EEG, especially relative power in the alpha band (8-13 Hz) and the theta-alpha ratio (TAR), defined as summed theta power divided by summed alpha power. Relative power is band power divided by total 1-100 Hz power. The plan/movement split is derived from surface EMG movement onset, so the one second before motion captures stimulus processing and the one second after captures execution. Lower relative alpha and higher TAR are treated as engagement markers, and TAR computed separately on the first and last third of trials tracks how engagement changes over time.","core_discovery":"The central claim is that EEG biomarkers, particularly relative alpha power, show statistically significant improvements in engagement during the multi-modal matching task compared with the motor-only task. The strongest evidence comes from the planning phase, where relative alpha power is significantly lower and the theta-alpha ratio significantly higher for the matching task across most brain regions, and the theta-alpha ratio remains stable over trials. The authors interpret these changes as greater and more sustained engagement, and they conclude that a robot task integrating image, word, and movement demands can keep users engaged for longer therapy sessions.","pith_inferences":["A direct extension would compare the matching task with a version where the image and spoken word are irrelevant to the movement, isolating task-relevant cognitive processing from mere sensory stimulation.","Because relative alpha power is normalized by total 1-100 Hz power, recomputing the analysis with absolute band power would show whether the effect reflects true alpha suppression or a shift in slower bands.","The stable theta-alpha ratio over trials could be developed into a control signal for adaptive assist-as-needed controllers that increase difficulty when engagement drops.","Patient populations may respond differently, since stroke survivors with language or motor deficits could find the combined task more demanding rather than more engaging."],"forward_implications":["Multi-modal matching tasks could support longer robotic rehabilitation sessions, because theta-alpha ratio stayed level across 100 trials instead of dropping as it did in the motor-only task.","EEG engagement markers could be monitored continuously during therapy, replacing subjective questionnaires that interrupt the session.","The planning-phase effect localizes most of the engagement gain to stimulus processing, so task design should emphasize the moment when users compare visual and auditory information.","If engagement indeed promotes neuroplasticity, this task format is a promising candidate for combined motor and language rehabilitation in stroke and aphasia populations."],"supporting_citations":[{"why":"Supplies the image-word matching paradigm that the paper adapts into its multi-modal robot task.","marker":"[23]"},{"why":"Prior evidence linking decreased alpha power to engagement in video lectures, used to interpret alpha decreases as engagement.","marker":"[26]"},{"why":"Prior work connecting concentration and immersion to EEG theta/alpha changes, used to justify the theta-alpha ratio.","marker":"[27]"},{"why":"Prior evidence that alpha power is modulated by attentional changes and virtual-reality immersion, supporting occipital-parietal interpretation.","marker":"[29]"},{"why":"Validation of single-channel EEG engagement measurement in virtual rehabilitation, used to support alpha-based engagement markers in rehab settings.","marker":"[31]"},{"why":"Prior evidence that delta oscillations increase during cognitive processing, used to interpret higher relative delta power as engagement.","marker":"[33]"},{"why":"Prior use of alpha/theta ratio neurofeedback for attention, used as support for the theta-alpha ratio marker.","marker":"[35]"},{"why":"Prior use of theta/alpha ratio to quantify visual-spatial attention, used as support for the theta-alpha ratio marker.","marker":"[36]"}],"fun_headline_variants":["EEG shows multimodal robot tasks sustain engagement longer","Multimodal robot tasks beat motor-only in EEG engagement","EEG: multimodal robot therapy holds engagement over time","Sustained engagement via multimodal robot tasks seen in EEG"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that lower relative alpha power and a higher theta-alpha ratio are valid markers of engagement, a mapping taken from prior literature and applied to every result without an independent engagement check in this study.","fun_headline_variants_meta":{"raw":{"variants":["EEG shows multimodal robot tasks sustain engagement longer","Multimodal robot tasks beat motor-only in EEG engagement","EEG: multimodal robot therapy holds engagement over time","Sustained engagement via multimodal robot tasks seen in EEG"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000166,"raw_usage":{"total_tokens":1240,"prompt_tokens":920,"completion_tokens":320,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":536,"completion_tokens_details":{"reasoning_tokens":256}},"tokens_in":536,"tokens_out":320,"duration_ms":3774,"temperature":1.0,"reasoning_tokens":256,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T11:03:28.473825+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A decisive test would run the same protocol while measuring absolute (non-normalized) alpha power plus a subjective engagement rating or behavioral performance metric; if absolute alpha does not differ between tasks, or if perceived engagement fails to track the EEG differences, the central claim would not be supported.","supporting_citations":[{"cited_title":"Transcranial direct current stimulation vs sham stimulation to treat aphasia after stroke: a randomized clinical trial,","cited_arxiv_id":null,"evidence_quote":"Supplies the image-word matching paradigm that the paper adapts into its multi-modal robot task."},{"cited_title":"Detecting fluctuations in student engagement and retention during video lectures using electroencephalography,","cited_arxiv_id":null,"evidence_quote":"Prior evidence linking decreased alpha power to engagement in video lectures, used to interpret alpha decreases as engagement."},{"cited_title":"Comparison between concentration and immersion based on eeg analysis,","cited_arxiv_id":null,"evidence_quote":"Prior work connecting concentration and immersion to EEG theta/alpha changes, used to justify the theta-alpha ratio."},{"cited_title":"Eeg alpha power is modulated by attentional changes during cognitive tasks and virtual reality immersion,","cited_arxiv_id":null,"evidence_quote":"Prior evidence that alpha power is modulated by attentional changes and virtual-reality immersion, supporting occipital-parietal interpretation."},{"cited_title":"Single-channel eeg measurement of engagement in virtual rehabilitation: A validation study,","cited_arxiv_id":null,"evidence_quote":"Validation of single-channel EEG engagement measurement in virtual rehabilitation, used to support alpha-based engagement markers in rehab settings."},{"cited_title":"The functional significance of delta oscillations in cog- nitive processing,","cited_arxiv_id":null,"evidence_quote":"Prior evidence that delta oscillations increase during cognitive processing, used to interpret higher relative delta power as engagement."},{"cited_title":"Alpha/theta ratio neurofeedback training for attention enhancement in normal developing children: A brief report,","cited_arxiv_id":null,"evidence_quote":"Prior use of alpha/theta ratio neurofeedback for attention, used as support for the theta-alpha ratio marker."},{"cited_title":"Index of theta/alpha ratio to quantify visual- spatial attention in dyslexics using electroencephalogram,","cited_arxiv_id":null,"evidence_quote":"Prior use of theta/alpha ratio to quantify visual-spatial attention, used as support for the theta-alpha ratio marker."}],"review_version":1}