2026-06-13·Daily digest

🔬 Research Pulse

Daily Digest

June 13, 2026

🤖 AI

🧠 LLMs

1. SpatialClaw: Rethinking Action Interface for Agentic Spatial Reasoning

Authors: Seokju Cho, Ryo Hachiuma, Abhishek Badki... Published: 2026-06-11 | Citations: 0 arXiv | PDF

Research Question: How does the design of the action interface (how tools are invoked) shape a tool-augmented VLM agent's capacity for open-ended 3D/4D spatial reasoning, given that existing single-pass code execution and structured tool-call interfaces both limit flexibility?

Summary: SpatialClaw is a training-free agentic framework that reframes spatial reasoning as iterative code execution in a stateful Python kernel preloaded with frames and perception/geometry primitives. By letting the VLM write one cell per step conditioned on prior outputs, it outperforms prior spatial agents by +11.2 points (59.9% avg) across 20 3D/4D benchmarks, with consistent gains across six VLM backbones.

Key Results: SpatialClaw achieves 59.9% average accuracy across 20 spatial reasoning benchmarks (covering static and dynamic 3D/4D tasks), outperforming the prior best spatial agent by +11.2 points. Gains are consistent across six VLM backbones from two model families with no benchmark- or model-specific adaptation, demonstrating that interface design (not model tuning) drives the improvement.

Key Findings:

Action interface design is a first-class bottleneck for spatial reasoning agents, independent of the underlying VLM
A stateful code interface beats both single-pass code execution and structured tool-call APIs for open-ended 3D/4D tasks
Improvements generalize across six VLMs from two families without model- or benchmark-specific tuning, indicating the gain is interface-driven

Technical Novelty: Using code as a stateful action interface for spatial reasoning agents: a persistent Python kernel pre-loaded with frames and perception/geometry primitives, where the VLM writes one executable cell per step conditioned on all prior text and visual outputs — enabling free composition of operations versus rigid structured tool-calls or one-shot code generation.

What's New: Prior spatial agents commit to a full plan via one-shot code or use rigid structured tool-calls; SpatialClaw is the first to combine a persistent stateful kernel, multimodal intermediate observations, and per-step code authoring as the unified action interface for spatial reasoning.

Extension Opportunities:

Add learned/RL-based policies on top of the stateful kernel to optimize which primitives to call, rather than relying purely on training-free VLM prompting
Extend the primitive suite to embodied/robotics settings (manipulation, navigation) where 4D spatial reasoning must close the loop with physical actions
Introduce caching, memoization, or sub-agent decomposition over the Python kernel state to scale to long-horizon video and multi-scene reasoning tasks

Replicability: Abstract does not mention code/data release. Training-free design means no GPU training cost; reproduction needs API/inference access to six VLM backbones plus compute to run perception/geometry primitives across 20 benchmarks — moderate inference budget, likely feasible on a single workstation with GPU for perception modules.

Research Gaps:

No evaluation in closed-loop embodied or robotic settings where actions affect the scene
Reliance on prompting a frozen VLM leaves open whether fine-tuning or RL on the code-action trajectory would compound the gains

2. Automated reproducibility assessments in the social and behavioral sciences using large language models

Authors: Tobias Holtdirk, Pietro Marcolongo, Anna Steinberg Schulten... Published: 2026-06-11 | Citations: 0 arXiv | PDF

Research Question: Can large language models automate reproducibility assessments in the social and behavioral sciences, replacing or augmenting resource-intensive human reanalysis that doesn't scale?

Summary: The authors build an LLM-driven pipeline that autonomously reanalyzes published social/behavioral studies and assesses whether original findings replicate. Across 76 studies it matches original effect sizes 41% of the time and original qualitative conclusions 96% of the time — beating human reanalysts (34%/74%) and suggesting LLMs can serve as a scalable infrastructure for systematic reproducibility auditing.

Key Results: On N=76 published social/behavioral studies with predefined claims, the LLM pipeline failed to produce viable effect sizes on 7 studies. On the remaining 69, it recovered original effect sizes within +/-0.05 Cohen's d tolerance in 41% of studies and matched the original qualitative conclusion in 96% of cases. Human reanalysts on the same task achieved only 34% effect-size recovery and 74% qualitative agreement — meaning LLMs outperformed humans on both metrics.

Key Findings:

LLM pipeline matched original qualitative conclusions in 96% of cases vs. 74% for human reanalysts
LLM recovered original effect sizes within +/-0.05 Cohen's d in 41% of studies vs. 34% for humans
Pipeline failed to produce any viable effect size on 7 of 76 studies (~9%), defining a clear coverage gap

Technical Novelty: Prior LLM-for-science work focused on summarization, code generation, or peer-review assistance. This is among the first end-to-end pipelines that ingests a paper + raw data, performs the statistical reanalysis autonomously, computes effect sizes, and renders a reproducibility verdict — benchmarked head-to-head against trained human reanalysts on a curated multi-study corpus.

What's New: First demonstration that LLMs can perform end-to-end automated reproducibility assessment — not just summarization or code suggestions — and quantitatively outperform human reanalysts on both effect-size recovery and qualitative verdicts in a curated multi-study benchmark.

Extension Opportunities:

Extend the pipeline to other empirical domains (biomedical RCTs, economics, ecology) where preregistered claims and open data exist, benchmarking against existing reproducibility consortia like the Reproducibility Project
Build a continuous auditing service that ingests new preprints/published papers with linked data and flags non-reproducible claims at submission or review time for journals
Investigate failure modes on the 7 'no viable effect size' studies and the qualitative-disagreement cases to design hybrid human-in-the-loop triage, where LLMs handle high-confidence cases and route ambiguous ones to humans

Replicability: The abstract does not state code/data availability. The N=76 study corpus likely comes from an existing reproducibility project (e.g., SCORE/DARPA or similar) and would need to be obtained. Compute is modest — LLM API calls plus lightweight statistical execution; reproducible on a single workstation with API access.

Research Gaps:

Unclear how performance generalizes beyond behavioral/social sciences to domains with more complex statistical models (multilevel, causal, time-series)
No analysis of why 41% effect-size recovery is so much lower than 96% qualitative agreement — suggests LLMs may converge on correct directional conclusions via different analytic paths, which has implications for trust and auditability

3. Agents-K1: Towards Agent-native Knowledge Orchestration

Authors: Zongsheng Cao, Bihao Zhan, Jinxin Shi... Published: 2026-06-11 | Citations: 0 arXiv | PDF

Research Question: How can LLM-based research agents move beyond shallow citation-graph orchestration (abstracts, surface mentions, flat cites edges) to orchestrate the deeper scientific knowledge — entities, claims, evidence, mechanisms, and method lineages — needed for genuine scientific reasoning?

Summary: Agents-K1 is an end-to-end pipeline that converts raw scientific papers into agent-native knowledge graphs using a multimodal full-paper parser, a 4B GRPO-trained extraction model, and a tri-source CLI for agent retrieval. Applied to 2.46M papers across six fields, it produces Scholar-KG (1M-paper subset released) and reportedly outperforms baselines on extraction, KG construction, and multi-hop scientific reasoning.

Key Results: The authors built Agents-K1 and ran the pipeline over 2.46 million scientific papers across six subjects to produce Scholar-KG, with a 1M-paper subset released publicly. They report superior performance (vs unspecified baselines) on scientific information extraction, knowledge graph construction, and multi-hop scientific reasoning benchmarks. The IE backbone is a 4B model trained with GRPO under a rule-based reward; no specific numeric scores are given in the abstract.

Key Findings:

Reducing papers to abstracts and flat citation edges loses the entities, claims, mechanisms, and method lineages required for scientific reasoning
A small 4B IE model with GRPO and rule-based rewards is sufficient to extract structured scientific knowledge at 2.46M-paper scale
Unifying web search, multimodal graph retrieval, and cross-document traversal in one CLI improves multi-hop scientific reasoning over single-source retrieval

Technical Novelty: Three coupled novelties: (1) a five-module parser schema that captures full-paper entities, multimodal evidence, citations, and typed inter-entity relations rather than abstract-only flat cites; (2) a compact 4B IE backbone trained with GRPO using rule-based rewards rather than relying on giant frontier models for extraction; (3) the graphanything CLI as a tri-source agent interface unifying web search, multimodal graph retrieval, and cross-document traversal.

What's New: Prior research-agent work focuses on agent orchestration over shallow citation graphs; Agents-K1 instead orchestrates structured scientific knowledge (typed relations, multimodal evidence, claims) extracted from full paper bodies, paired with a purpose-trained compact extractor and a tri-source retrieval interface.

Extension Opportunities:

Apply the five-module multimodal parser + GRPO-trained IE backbone to non-scientific corpora (legal, clinical, patent) since the authors note the pipeline generalizes to general-domain text and schema-conformant synthesis
Build domain-specific reasoning agents on top of the released 1M-paper Scholar-KG subset using the graphanything CLI's tri-source interface (web + multimodal graph retrieval + cross-document traversal)
Extend the schema with temporal/provenance edges to track method-lineage evolution over time, enabling agents to detect when claims are superseded or refuted

Replicability: Partial — a 1M-paper subset of Scholar-KG is released and the full KG is accessible via an SCP link. The abstract does not confirm release of model weights, training code, or the parser. Reproducing the corpus build would need substantial compute (PDF parsing + GRPO training of a 4B model + extraction over 2.46M papers); likely multi-GPU cluster scale.

Research Gaps:

Existing scientific KGs reduce papers to abstracts and flat cites edges, omitting entities, claims, evidence, and method lineages
Research agents lack a unified retrieval interface that combines live web search with multimodal graph and cross-document traversal

🦾 ROBOTICS

1. LabVLA: Grounding Vision-Language-Action Models in Scientific Laboratories

Authors: Baochang Ren, Xinjie Liu, Xi Chen... Published: 2026-06-11 | Citations: 0 arXiv | PDF

Research Question: How can Vision-Language-Action (VLA) models be adapted to execute scientific laboratory protocols, given that existing policies are trained on household/tabletop demonstrations and lack exposure to lab instruments, transparent liquids, and protocol workflows across diverse robot embodiments?

Summary: LabVLA adapts Vision-Language-Action models to scientific laboratory protocol execution by addressing two bottlenecks: data scarcity and embodiment diversity. It introduces RoboGenesis, a simulation data engine producing validated multi-embodiment lab demonstrations, and trains a Qwen3-VL-4B-Instruct backbone with a two-stage recipe (FAST action-token pretraining, then flow-matching with a DiT action expert under knowledge insulation), achieving state-of-the-art on the LabUtopia benchmark.

Key Results: LabVLA achieves the highest average success rate among all evaluated baselines on the LabUtopia benchmark under both in-distribution and out-of-distribution settings. The paper demonstrates a working pipeline combining: (1) RoboGenesis simulation-based data engine generating validated demonstrations across robot profiles, and (2) a two-stage training recipe (FAST action token pretraining on Qwen3-VL-4B-Instruct + flow matching posttraining with DiT action expert under knowledge insulation). Specific success-rate numbers are not provided in the abstract.

Key Findings:

A two-stage training recipe (discrete FAST tokens first, then continuous flow-matching) outperforms direct continuous-control finetuning by making the VLM action-aware before attaching the action expert
Knowledge insulation during DiT action expert attachment preserves the pretrained VLM's reasoning while enabling continuous control
Simulation-generated, validated, multi-embodiment lab demonstrations are sufficient to achieve best-in-class success on both in-distribution and OOD lab tasks

Technical Novelty: Three combined novelties: (1) RoboGenesis — a simulation data engine that composes lab workflows from atomic skills with rollout validation/filtering and structured export across robot embodiments; (2) a two-stage training recipe where FAST discrete action token pretraining makes the VLM 'action aware' before continuous control; (3) flow-matching posttraining with a DiT action expert under 'knowledge insulation' to preserve the pretrained VLM backbone while learning continuous actions.

What's New: Unlike prior VLA work focused on household/tabletop manipulation, this paper targets the scientific lab domain with its unique challenges (transparent liquids, instruments, rigid protocols) and explicitly treats embodiment diversity as a first-class problem via a unified data engine and multi-profile export, paired with a novel discrete-then-continuous training recipe with knowledge insulation.

Extension Opportunities:

Extend RoboGenesis with additional atomic skills for specialized lab domains (e.g., biology wet-lab, materials synthesis, electrochemistry) and benchmark transfer to real hardware beyond simulation
Integrate LabVLA with LLM-based protocol planners (e.g., reading literature → generating protocols → executing) to close the loop from hypothesis to bench execution
Add multi-robot coordination and long-horizon protocol chaining across embodiments, evaluating robustness to transparent/reflective objects and liquid handling failures

Replicability: The abstract does not mention code/data release. Reproduction would require: Qwen3-VL-4B-Instruct backbone (open), a simulation environment supporting LabUtopia benchmark, multi-GPU training for two-stage pipeline (likely 8+ A100/H100 for FAST pretraining plus DiT flow-matching), and the RoboGenesis data engine. Without released artifacts, full replication is non-trivial.

Research Gaps:

No real-robot evaluation discussed in the abstract — sim-to-real transfer for transparent liquids and precision lab instruments remains open
Coverage of atomic skills and breadth of lab workflows (chemistry vs biology vs materials) and long-horizon multi-step protocols are not characterized

2. Low cost, easily manufactured, highly flexible strain and touch sensitive fiber for robotics applications

Authors: Christian Diaz Herrera, Srushti Raste, Simin Liu... Published: 2026-06-11 | Citations: 0 arXiv | PDF

Research Question: How can stretch and touch sensors for robotics be made affordable, fast to manufacture, and highly flexible without requiring expensive materials or specialized equipment?

Summary: The paper introduces a sub-dollar-per-foot conductive fiber sensor built from commodity conductive thread inside silicone tubing, assembled in two minutes with a $2 needle threader. It serves as both a resistive strain sensor and a capacitive touch/proximity sensor, validated across five robotics demos including pneumatic actuator pose estimation and hand-following — and is knittable and field-repairable.

Key Results: Demonstrated a conductive fiber built from off-the-shelf parts: conductive thread ($0.07/ft), silicone tubing ($0.94/ft), and a $2 loop-style needle threader, with a 20 cm length manufactured in 2 minutes. Validated across five applications: resistive strain sensing for (1) triggering grasp in a pneumatic assistive finger, (2) sensing pose of a pneumatic robotic strap, (3) estimating pose of a flexible solid; and capacitive sensing for (4) touch-triggered robot arm motion and (5) near-field hand-following. Capacitive variants were knitted to showcase flexibility, and a cut-fiber repair method was demonstrated.

Key Findings:

A 20 cm fiber can be fabricated in 2 minutes from off-the-shelf parts costing roughly $0.20 in materials, dramatically undercutting typical soft-sensor BOMs
The same fiber works in both resistive (strain) and capacitive (touch/near-field) modes, validated across five distinct robotic applications
The fiber is flexible enough to be knitted into fabric and can be repaired after being cut, addressing two practical pain points for soft-robotics deployment

Technical Novelty: Combines commodity conductive thread inside silicone tubing using a $2 needle threader — eliminating the specialized equipment (e.g., liquid metal injection, screen printing, CNT coating) typical of prior soft sensor fabrication, while supporting both resistive and capacitive modalities and being knittable and repairable.

What's New: Unlike prior soft sensors that rely on liquid metals, custom inks, or specialized fabrication, this design uses only commodity sewing and tubing components with hand tools, while still demonstrating dual-mode sensing, knittability, and repairability.

Extension Opportunities:

Integrate the fiber into wearable haptic gloves or soft exosuits, leveraging knittability to map distributed strain/touch signals for teleoperation
Develop closed-loop control for soft pneumatic robots using arrays of these fibers as proprioceptive feedback, with ML-based calibration to handle nonlinearity and drift
Scale manufacturing via automated threading machines or braiding looms and benchmark cost/throughput tradeoffs against commercial sensor yarns

Replicability: Highly replicable from a hardware standpoint — all materials are commercial off-the-shelf with stated prices, and the threader-based assembly is described concretely. No compute requirements for fabrication; reproducing the robotic demos requires standard pneumatic actuators and a commercial robot arm. Abstract does not mention released code, datasets, or CAD files.

Research Gaps:

Quantitative characterization of sensor accuracy, hysteresis, drift, durability, and signal-to-noise versus established soft sensors is not summarized in the abstract
Manufacturing scalability discussion is acknowledged but not yet demonstrated at production volume

3. WT-UMI: Tactile-based Whole-Body Manipulation via Force-Supervised Contact-Aware Planning

Authors: Jaehwi Jang, Zhaoyuan Gu, Alfred Cueva... Published: 2026-06-11 | Citations: 0 arXiv | PDF

Research Question: How can humanoid robots perform whole-body manipulation of bulky, deformable, and shared-load objects when imitation policies typically treat contact force only implicitly, and how can complementary human demonstration (natural force) and teleoperation (executable actions) data be unified despite their inherent trade-offs?

Summary: WT-UMI is a wearable whole-body tactile interface that captures synchronized tactile images, contact forces, and end-effector poses from both human demonstrators and teleoperated humanoids, enabling a force-supervised policy that predicts both pose and contact-force trajectories. By learning a correction from teleoperation data to map natural human contact patterns into robot-executable targets, it explicitly closes the force-imitation gap and feeds predictions into a tactile admittance controller for contact-rich whole-body manipulation.

Key Results: Across five contact-rich tasks spanning deformable objects, bulky rigid objects, and human-humanoid collaboration, WT-UMI improves success rate and reduces contact-position tracking error compared to four policy baselines. Specific numerical results require consulting the full paper; the abstract reports qualitative improvements without exact percentages.

Key Findings:

Unified human-and-robot tactile data collection via a single wearable interface reduces the modality mismatch that typically forces a choice between natural forces (human) and executable actions (teleop)
Explicitly supervising contact-force trajectories alongside pose chunks outperforms baselines that handle force only implicitly on contact-rich tasks
A learned pose-correction module conditioned on force is sufficient to translate human kinematics into contact-aware humanoid targets across deformable, bulky-rigid, and shared-load tasks

Technical Novelty: A wearable whole-body tactile interface (WT-UMI) usable in both human-demonstration and humanoid-teleoperation modes, combined with (1) a force-conditioned target-pose correction module that translates human poses into contact-aware robot targets via teleoperation-learned corrections, and (2) a force-supervised planner that jointly predicts pose chunks and contact-force trajectories used as the reference for a tactile-based admittance controller — explicitly closing the human-to-robot force-imitation gap.

What's New: Unlike prior UMI-style interfaces that capture only pose/vision, and unlike imitation policies that treat force implicitly, WT-UMI explicitly couples whole-body tactile sensing with force-supervised planning and a teleoperation-grounded correction module — bridging the human-demo vs. teleop trade-off rather than picking one.

Extension Opportunities:

Extend the wearable tactile interface to multi-agent humanoid collaboration where two robots coordinate force-supervised planning over a shared object
Replace the admittance controller with a learned impedance policy that adapts stiffness based on predicted force trajectories for more compliant interaction
Apply the force-conditioned target-pose correction module to dexterous bimanual tasks (e.g., garment folding, assembly) where deformation feedback drives in-hand regrasping

Replicability: A project page exists at https://wt-umi.github.io/WTUMI/, but code/dataset release is not confirmed in the abstract. Reproducing would require a humanoid platform, custom wearable tactile sensors (likely vision-based tactile arrays for the 'tactile images'), teleoperation setup, and standard imitation-learning compute (single-node multi-GPU sufficient for pose/force chunk prediction models).

Research Gaps:

Most imitation policies for whole-body manipulation treat contact force only implicitly, limiting performance on bulky/deformable/shared-load objects
Human demonstrations and teleoperation data each cover only half the needed signal (natural force vs. executable actions), with no prior interface unifying both modalities

💻 COMPUTE

1. Measurement-Based Performance Evaluation of SmartRSUs with Heterogeneous Antenna Architectures for V2X Communications

Authors: Marco Savarese, Gaetano Orazio Cauchi, Salvatore Iandolo... Published: 2026-06-11 | Citations: 0 arXiv | PDF

Research Question: How do different V2X antenna architectures (integrated all-in-one rooftop vs. external dual ITS-G5) on Smart Roadside Units impact real-world communication reliability, coverage, and latency in C-ITS deployments?

Summary: The paper empirically compares two custom SmartRSU designs — one with an integrated all-in-one GNSS+comm rooftop antenna and one with external dual ITS-G5 antennas plus a separate GNSS antenna — using a co-located commercial RSU as ground truth. Measurements of RSSI, packet loss, coverage, and bidirectional latency expose concrete trade-offs between integrated and external antenna placements for real-world C-ITS deployment.

Key Results: The authors built two custom SmartRSUs on a proprietary OBU platform and measured RSSI, packet loss, coverage, and bidirectional end-to-end latency (OBU-to-infrastructure and infrastructure-to-OBU) at 5.9 GHz (IEEE 802.11p / ITS-G5). A commercial COTS V2X RSU was co-located as ground-truth benchmark under identical environmental conditions, demonstrating measurable trade-offs between integrated and external antenna configurations.

Key Findings:

Antenna design and placement materially affect V2X communication reliability and latency in deployed RSUs
Integrated all-in-one rooftop modules and external dual-antenna configurations exhibit distinct trade-offs across coverage, RSSI, and packet loss
Co-locating a COTS RSU as benchmark enables objective validation of custom SmartRSU performance under identical conditions

Technical Novelty: Direct, co-located A/B measurement comparison of two custom SmartRSU antenna architectures (integrated rooftop module vs. external dual ITS-G5 + dedicated GNSS) using a shared proprietary OBU-derived infrastructure platform, validated against a COTS RSU baseline — most prior work evaluates a single RSU design or simulates antenna effects rather than empirically isolating them.

What's New: Combines a custom OBU-derived RSU platform with a controlled co-located COTS benchmark to isolate the effect of antenna architecture, rather than treating the RSU as a black box or relying on simulation.

Extension Opportunities:

Repeat the campaign with C-V2X PC5 (3GPP Rel-16/17) sidelink alongside ITS-G5 to compare antenna-architecture sensitivity across radio access technologies
Add mobility scenarios with multiple OBU-equipped vehicles at varying speeds and densities to characterize Doppler and multi-user effects beyond static infrastructure benchmarking
Integrate ML-based adaptive beam/antenna selection that switches between integrated and external arrays based on real-time RSSI/PER telemetry

Replicability: No code or dataset release is mentioned in the abstract. Reproduction requires custom OBU hardware adapted for RSU use, ITS-G5 transceivers at 5.9 GHz, GNSS modules, a COTS V2X RSU for benchmarking, and outdoor test site access — compute needs are minimal; hardware and spectrum access are the primary barriers.

Research Gaps:

No evaluation under high-mobility or dense multi-vehicle scenarios; results appear limited to infrastructure-side static testing
ITS-G5 / IEEE 802.11p only — no comparison against C-V2X PC5 or NR sidelink under the same antenna architectures

2. Observation of Non-Gaussian Magnon Dynamics in a Two-Dimensional Long-Range XY Model

Authors: S. -A. Guo, J. -Y. Tan, J. Ye... Published: 2026-06-11 | Citations: 0 arXiv | PDF

Research Question: How can one experimentally verify non-Gaussian magnon dynamics in a 2D long-range XY model while overcoming decoherence, statistical fluctuations, and parameter miscalibration that typically obscure such signatures?

Summary: The authors use a trapped-ion quantum simulator to observe the crossover from Gaussian to non-Gaussian magnon dynamics in a 2D long-range XY model. By comparing measured high-order spin correlations to mean-field and Holstein-Primakoff predictions in a manner robust to calibration errors, they provide a verifiable signature of dynamics beyond classically tractable approximations.

Key Results: Demonstrated a crossover between Gaussian and non-Gaussian dynamics on a 2D XY model with long-range, spatially structured interactions using a trapped ion quantum simulator. Verified single-spin observable dynamics across different magnon excitation densities, then compared high-order spin correlations against mean-field and Holstein-Primakoff (HP) approximations to show non-Gaussian behavior in a calibration-error-independent manner.

Key Findings:

A clear Gaussian-to-non-Gaussian crossover emerges as magnon excitation density is varied in the 2D long-range XY model
Single-spin observable dynamics match the engineered Hamiltonian, validating the simulator's fidelity
High-order correlations deviate from both mean-field and Holstein-Primakoff predictions in a way that survives calibration uncertainty

Technical Novelty: A calibration-error-independent comparison of high-order spin correlations against both mean-field and Holstein-Primakoff baselines on a 2D long-range XY model — prior work either focused on 1D, used Gaussian-only observables, or was confounded by miscalibration when probing non-Gaussianity.

What's New: Combines a 2D (not 1D) long-range XY platform with a calibration-independent witness for non-Gaussianity, bridging classically simulatable regimes (HP/mean-field) and potential quantum-advantage regimes within a single experiment.

Extension Opportunities:

Scale the trapped-ion platform to larger 2D lattices to probe regimes where HP and mean-field theories fully break down, mapping the boundary of classical simulability
Extend the protocol to other spatially structured interaction profiles (e.g., dipolar, frustrated geometries) to study non-Gaussian dynamics in spin liquids or topological phases
Develop calibration-independent witnesses for higher-order (4th, 6th) correlators to certify quantum advantage in near-term analog simulators

Replicability: No code/data link mentioned in the abstract. Reproduction requires a trapped-ion quantum simulator capable of engineering 2D long-range XY Hamiltonians with spatially structured couplings — high-end experimental physics infrastructure, not commodity compute.

Research Gaps:

Lacks quantitative bounds on where exactly classical methods fail vs. quantum simulation provides advantage
Open question whether the calibration-independent witness extends to other Hamiltonians or larger system sizes

3. Quantum optical photoelectron interferometry

Authors: Jonathan Dubois, Viviane Cotte, Richard Taïeb... Published: 2026-06-11 | Citations: 0 arXiv | PDF

Research Question: How do photon statistics of quantum light fields imprint onto photoelectron observables in multiphoton ionization, and specifically how do quantum-optical properties (squeezing, correlations) modify attosecond RABBIT spectroscopy signals?

Summary: The paper develops a general quantum-optical theory of multiphoton photoionization that directly maps photon autocorrelation and cross-correlation functions onto photoelectron spectra. Applying it to RABBIT spectroscopy, the authors show that sideband amplitude, contrast, and phase encode the quantum statistics of the driving light, validated analytically and numerically for a squeezed-coherent IR field.

Key Results: Derived an analytical framework showing autocorrelation and cross-correlation functions of the driving light field are directly mapped onto photoelectron spectra. Demonstrated that in RABBIT, the amplitude, contrast, and phase of sideband oscillations vs pump-probe delay encode the quantum nature of light. Validated analytical theory against numerical simulations for classical harmonics + squeezed-coherent IR field with 'excellent agreement.' Treated multiple regimes: correlated IR/harmonic modes, uncorrelated modes with non-classical harmonic statistics.

Key Findings:

Photon statistical functions (auto- and cross-correlations) are directly imprinted on photoelectron spectra in multiphoton processes
RABBIT sideband contrast and phase serve as observable signatures of non-classical light statistics and IR–harmonic correlations
Analytical theory matches numerical TDSE-style simulations with excellent agreement for classical harmonics + squeezed-coherent IR
The interplay between classical and quantum correlations governs photoemission coherence

Technical Novelty: First general theoretical framework linking photon autocorrelation/cross-correlation functions directly to photoelectron spectra in multiphoton ionization. Prior RABBIT theory assumed classical coherent fields; this work generalizes to arbitrary quantum statistics including squeezed and correlated multimode states, identifying contrast and phase of sidebands as quantum-optical observables.

What's New: Bridges quantum optics and attosecond science by treating the driving field as quantized rather than classical, recasting RABBIT as a quantum-light spectroscopy. Identifies sideband contrast/phase — previously interpreted purely in terms of electronic phase delays — as probes of photon statistics.

Extension Opportunities:

Simulate RABBIT with other non-classical light states (Fock, cat states, thermal) to map a state-discrimination protocol from sideband observables
Extend the formalism to molecular targets where vibrational/rotational degrees of freedom couple to photon statistics, enabling quantum-light molecular attosecond spectroscopy
Develop an inverse-problem pipeline: reconstruct unknown photon-statistics (g^(1), g^(2)) of an XUV/IR source directly from measured sideband contrast and phase

Replicability: No code or data availability mentioned in the abstract. Reproduction requires solving the TDSE for atomic targets driven by quantized light fields (squeezed coherent IR + harmonic comb) — feasible on a workstation or small cluster for single-active-electron atoms; standard attosecond/strong-field numerics expertise needed.

Research Gaps:

Experimental demonstration with genuinely non-classical XUV/harmonic sources is still pending
Framework focuses on atomic targets; molecular and condensed-matter extensions remain open

⚡ ENERGY

1. Mathematical Modeling of HDV RNA, HBV DNA, and HBsAg Dynamics during Lonafarnib-Based Therapy: Insights from the LOWR HDV-1 Study

Authors: Adquate Mhlanga, Louis Shekhtman, Rami Zakh... Published: 2026-06-11 | Citations: 0 arXiv | PDF

Research Question: How do HDV RNA, HBV DNA, and HBsAg kinetics evolve during Lonafarnib-based therapy in HBV/HDV coinfected patients, and what mechanistic model explains the observed biphasic decline, viral breakthrough, and HBV rebound phenomena?

Summary: The authors develop a mechanistic mathematical model fit to viral kinetic data from 15 HBV/HDV coinfected patients receiving Lonafarnib-based therapy in the LOWR HDV-1 trial, quantifying HDV RNA half-life (~1.26 days) and treatment efficacy (~94%). The model explains the biphasic HDV decline via time-dependent efficacy and the surprising rebound in HBV DNA via a hypothesis that HDV normally suppresses HBV production below an inhibitory threshold.

Key Results: Modeled kinetics in 15 coinfected patients from LOWR HDV-1 study. Measured HDV RNA serum half-life of 1.26 days (95% CI: 1.05-1.47), treatment efficacy of 94% in inhibiting HDV RNA production across all treatments (95% CI: 89%-97%), with 2nd-phase efficacy reaching 98.9%. Median 4-fold (IQR: 1-28) increase in HBV DNA production rate when HDV declined below an inhibitory threshold. HDV decline begins after 0-2 day delay.

Key Findings:

HDV RNA has a short serum half-life of 1.26 days, indicating rapid turnover and informing dosing strategy
LNF monotherapy often led to flat-partial-response and viral breakthrough, while combination with ritonavir or PEG-IFNα produced sustained biphasic HDV decline
HBV DNA paradoxically increases during HDV suppression (median 4-fold), suggesting HDV normally inhibits HBV replication — a clinically important interaction
HBsAg remained stable across treatments, consistent with a constant pool of HBsAg-producing cells unaffected by short-term antiviral therapy

Technical Novelty: First mathematical model to jointly capture HDV RNA, HBV DNA, and HBsAg kinetics under LNF-based regimens, introducing a time-dependent efficacy term to explain biphasic HDV decline and an HDV-dependent inhibitory threshold mechanism to explain paradoxical HBV DNA rebound during HDV suppression.

What's New: Prior HDV kinetic models focused on interferon-based therapy; this work uniquely leverages LNF's HDV-specific mechanism to dissect HDV-HBV viral interference and introduces both time-dependent efficacy and threshold-dependent HBV reactivation as novel modeling constructs.

Extension Opportunities:

Extend the model to incorporate newer HDV therapies like bulevirtide (entry inhibitor) to compare mechanistic efficacy profiles and predict combination regimens
Build a predictive clinical decision-support tool that identifies patients likely to experience viral breakthrough on LNF monotherapy vs combination therapy based on early kinetic markers
Couple the model with intracellular HBV/HDV replication dynamics to model HBsAg-producing cell turnover and predict long-term functional cure probability

Replicability: No code or data availability explicitly mentioned in abstract. Reproduction would require access to LOWR HDV-1 clinical trial patient kinetic data (15 patients). Compute requirements are minimal — ODE-based viral dynamics models run trivially on standard CPUs using tools like MATLAB, R (deSolve), or Python (scipy).

Research Gaps:

Model does not capture intrahepatic dynamics (cccDNA, intracellular HDV RNA) — limits inference about functional cure
Small cohort (n=15) and lack of long-term follow-up beyond LNF treatment window limit generalizability and prediction of post-treatment relapse

2. Micron-sized magnonic 3-port rectilinear circulator

Authors: Kevin S. Weber, Loic Temdie, Vincent Castel... Published: 2026-06-11 | Citations: 0 arXiv | PDF

Research Question: How can non-reciprocal microwave components (circulators) be miniaturized to micron scale and made compatible with integrated circuit architectures, overcoming the bulky size limitations of conventional ferrite-based circulators?

Summary: The authors demonstrate the first micron-scale 3-port magnon circulator using nanowire grating arrays that chirally excite unidirectional spin-wave beams along three rectilinear channels. Operating in the 2-8 GHz band with magnetic field tunability up to 100 mT, this device offers a path toward integrated, miniaturized non-reciprocal microwave components.

Key Results: First experimental demonstration of a micron-sized 3-port magnon circulator using nanowire grating arrays for chiral spin-wave excitation. Full 3-port spin-wave spectroscopy confirmed genuine circulation between three rectilinear channels. Operation tunable across 2-8 GHz via external magnetic field up to 100 mT and grating geometry controlling wavevectors.

Key Findings:

Genuine 3-port spin-wave circulation experimentally verified via full S-parameter spectroscopy at micron scale
Operating frequency tunable across the entire 2-8 GHz range via bias field (≤100 mT) and grating geometry
Nanowire gratings successfully produce chiral, unidirectional spin-wave excitation suitable for rectilinear channel routing

Technical Novelty: The original rectilinear 3-channel geometry leveraging chiral spin-wave excitation from nanowire gratings—prior magnonic non-reciprocal proposals were largely theoretical or used circular/Y-junction topologies; this is the first experimental micron-scale realization with three straight unidirectional spin-wave beams.

What's New: Combines chiral magnonic excitation via nanowire gratings with a novel rectilinear three-channel topology to achieve the first experimental micron-scale magnon circulator, departing from bulky ferrite junction circulators and prior purely theoretical magnonic designs.

Extension Opportunities:

Integrate the magnonic circulator with on-chip CMOS/superconducting qubit readout lines to replace bulky ferrite circulators in quantum computing dilution refrigerators
Engineer broadband operation by cascading multiple gratings with staggered periods or designing chirped nanowire arrays to overcome the narrow-band limitation
Explore field-free operation using exchange-biased or synthetic antiferromagnet layers beneath the YIG film to eliminate the external bias magnet requirement

Replicability: Abstract does not mention code/data availability. Reproduction requires: high-quality YIG thin film (likely LPE-grown), e-beam lithography for nanowire gratings, vector network analyzer for spin-wave spectroscopy, electromagnet (up to 100 mT), and microwave probe station — standard for a magnonics lab but non-trivial fabrication.

Research Gaps:

Narrow operational bandwidth limits use in broadband communication systems
Requires external bias field up to 100 mT, complicating integration without permanent magnets or self-biased materials

3. Bandedge-state-limited single-photon emission from volumetric quantum design of 2D colloidal quantum wells

Authors: Xiao Liang, Bo Wang, Yue Yu... Published: 2026-06-11 | Citations: 0 arXiv | PDF

Research Question: How can solution-processable single-photon sources overcome the particle-to-particle heterogeneity of 3D colloidal quantum dots while simultaneously resolving the long-standing problems of multiexciton emission and surface-state susceptibility that have prevented 2D colloidal quantum wells (CQWs) from serving as reliable single-photon emitters?

Summary: The paper introduces volumetric quantum design (VQD) of 2D colloidal quantum wells to deliver reliable room-temperature single-photon emission, overcoming the multiexciton-emission and surface-susceptibility problems that had previously disqualified CQWs as quantum-light sources. VQD-CQWs achieve near-blinking-free emission (>99.5% on-time), strong antibunching (g²(0) = 0.041), and high linear polarization (73%) without cavities, establishing CQWs as a homogeneous, scalable alternative to colloidal quantum dots.

Key Results: The authors demonstrate room-temperature single-photon emission from volumetric-quantum-design CQWs (VQD-CQWs) with: (1) near-blinking-free behavior with on-time >99.5%, (2) fluence-insensitive antibunching with g²(0) = 0.041, (3) linear polarization up to 73% under cavity-free conditions, and (4) a highly localized single bandedge state achieved via lateral confinement within the exciton coherent area combined with a thick, strain-relieved quantum-barrier shell that decouples bandedge excitons from surface states.

Key Findings:

VQD laterally confines bandedge excitons within the exciton coherent area, creating a bandedge-state-filling bottleneck that suppresses multiexciton emission and yields fluence-insensitive antibunching (g²(0) = 0.041)
A thick, strain-relieved quantum-barrier shell decouples emitting states from surface defects, achieving on-time >99.5% (near-blinking-free) at room temperature
Synergistic transition-dipole and electric-field anisotropies in the 2D geometry produce up to 73% linear polarization under cavity-free conditions, a property difficult to obtain from 0D quantum dots without external structuring

Technical Novelty: The volumetric quantum design (VQD) approach is new: it simultaneously achieves (a) lateral confinement of bandedge excitons within the exciton coherent area, suppressing multiexciton emission via a bandedge-state-filling bottleneck, and (b) vertical decoupling from surface states using a thick, strain-relieved quantum-barrier shell — overcoming the conventional thickness-vs-confinement trade-off in CQWs that prior core/shell or core-only architectures could not solve.

What's New: Prior CQW work struggled with the thickness-confinement trade-off and surface-state coupling, making single-photon emission unreliable. This paper is the first to demonstrate that engineered lateral + vertical confinement (VQD) can produce a single, localized bandedge state in a 2D colloidal platform, leveraging the inherent atomic-flatness uniformity of CQWs that 3D quantum dots fundamentally lack.

Extension Opportunities:

Integrate VQD-CQWs into photonic cavities or waveguides to boost extraction efficiency and indistinguishability for on-chip quantum networking applications
Extend the volumetric quantum design strategy to other 2D semiconductor material systems (e.g., perovskite nanoplatelets, II-VI alloys) to engineer tunable single-photon wavelengths across visible/NIR
Develop scalable assembly/printing techniques exploiting the inherent uniformity of VQD-CQWs to build deterministic single-photon arrays for quantum imaging or multiplexed QKD

Replicability: No mention of code/data availability in the abstract. Reproduction would require colloidal synthesis infrastructure (Schlenk line, glovebox, precursor chemistry expertise), single-particle spectroscopy setup (HBT interferometer, confocal microscope, time-correlated single-photon counting), and structural characterization (HRTEM, XRD). Compute requirements are modest — primarily for k·p or DFT modeling of confinement; the experimental burden dominates.

Research Gaps:

Indistinguishability and Hong-Ou-Mandel interference of photons from VQD-CQWs are not reported — critical for linear-optical quantum computing
Scalable deterministic positioning and integration of these emitters into photonic circuits remains unaddressed

🔬 MATERIALS

1. Band-Selective Tunneling and Anisotropic Multiband Superconductivity in V$_2$Ga$_5$

Authors: Jozef Haniš, Jozef Kačmarčík, Filip Košuth... Published: 2026-06-11 | Citations: 0 arXiv | PDF

Research Question: How do directional tunneling experiments in anisotropic multiband superconductors produce apparent reversals of gap hierarchies, and how can these be reconciled with bulk thermodynamic measurements within a self-consistent multiband pairing framework?

Summary: The paper combines DFT, AC calorimetry, and directional scanning tunneling spectroscopy on tetragonal V2Ga5 (Tc ≈ 3.5 K) to build a self-consistent anisotropic multiband s-wave (A1g) pairing model. It shows that an apparent reversal of gap hierarchies between tunneling directions arises from band-selective tunneling shaped by Fermi velocity hot spots and Fermi surface topology, not from intrinsic gap magnitudes.

Key Results: The authors characterized tetragonal V2Ga5 (Tc ≈ 3.5 K) using AC calorimetry, directional low-T STS, and DFT. They constructed a self-consistent anisotropic multiband singlet s-wave model in the A1g representation that simultaneously fits specific heat and upper critical field Hc2 anisotropy, and demonstrated that direction-dependent tunneling gap hierarchies invert relative to bulk gaps due to band-selective tunneling weighted by Fermi velocity 'hot spots' and Fermi surface topology.

Key Findings:

V2Ga5 is well described by a fully symmetric A1g anisotropic multiband singlet s-wave pairing state that simultaneously reproduces specific heat and upper critical field anisotropy
Directional STS shows an apparent reversal of gap hierarchy that is not a property of the bulk gaps but a consequence of band-selective tunneling
Localized Fermi velocity hot spots combined with Fermi surface topology determine which bands dominate the tunneling signal along a given crystallographic direction

Technical Novelty: The explicit demonstration that directional tunneling gap hierarchies can invert relative to bulk thermodynamic gaps due to Fermi-velocity 'hot spots' modulating tunneling matrix elements — combined with a self-consistent anisotropic A1g multiband model that jointly fits Cp(T) and Hc2 anisotropy in V2Ga5.

What's New: Prior multiband STS interpretations typically map tunneling peaks directly to bulk gap magnitudes; this work introduces a microscopic, geometry-aware band-selectivity mechanism (Fv hot spots × FS topology) that explains why directional STS can invert apparent gap order, while remaining consistent with bulk thermodynamics.

Extension Opportunities:

Apply the band-selective tunneling framework (Fermi-velocity-weighted projection) to reinterpret existing directional STS data on other anisotropic multiband superconductors such as MgB2, FeSe, or 2H-NbSe2
Develop an open-source toolkit that takes DFT Fermi surfaces + gap functions and predicts orientation-resolved tunneling conductance, enabling automated screening of candidate multiband superconductors
Extend the A1g multiband model to include spin-orbit coupling or non-s-wave admixtures to test whether V2Ga5 hosts subleading unconventional pairing channels detectable via field-angle-resolved calorimetry

Replicability: No code or data availability is mentioned in the abstract. Reproduction would require: a dilution-refrigerator STM with directional tip access on oriented single crystals, AC calorimetry capability, and modest DFT compute (standard plane-wave DFT + Fermi surface integration, feasible on a small HPC cluster).

Research Gaps:

No quantitative framework yet generalizes the Fermi-velocity-weighted tunneling correction across broader classes of multiband superconductors
Direct phase-sensitive confirmation of the A1g assignment (vs. possible sign-changing s± components) is not established

2. Andreev Reflection to Probe Momentum-Dependent Spin Polarization in Altermagnet CrSb

Authors: Yan Zhang, Yixuan Luo, Yue Yang... Published: 2026-06-11 | Citations: 0 arXiv | PDF

Research Question: How can one experimentally verify the predicted momentum-dependent (k-dependent) spin polarization in altermagnetic CrSb, distinguishing it from conventional antiferromagnets, and resolve its real-space domain structure?

Summary: The authors use mechanical point-contact Andreev reflection spectroscopy with superconducting Nb tips on CrSb single crystals to directly measure momentum-dependent spin polarization (61.9-73.4%) along three crystallographic orientations, providing direct experimental evidence for g-wave altermagnetism. Spatial line-scans further resolve altermagnetic domains of 250-500 nm separated by ~250 nm domain walls, establishing Andreev reflection as a new diagnostic for k-dependent spin textures.

Key Results: Using mechanical point-contact spectroscopy (MPCS) with superconducting Nb tips, the authors measured Andreev reflection on CrSb single crystals along three crystallographic orientations, extracting distinct spin polarizations: ~73.4% for (0001), ~67.9% for (-1-120), and ~61.9% for (10-10). Spatial line-scan conductance spectra revealed altermagnetic domains 250-500 nm in size separated by ~250 nm wide domain walls.

Key Findings:

Spin polarization in CrSb is strongly anisotropic across crystallographic planes: 73.4% on (0001), 67.9% on (-1-120), and 61.9% on (10-10), directly consistent with g-wave altermagnetic spin splitting.
Altermagnetic domains in CrSb have a characteristic size of 250-500 nm with domain walls ~250 nm wide, resolvable via spatial variation of point-contact conductance spectra.
Andreev reflection spectroscopy can serve as a momentum-selective probe of k-dependent spin textures, complementing ARPES and neutron techniques.

Technical Novelty: Prior probes of altermagnetic spin splitting relied on ARPES, neutron scattering, or transport anomalies. This work is the first to use Andreev reflection spectroscopy with superconducting tips as a direct, momentum-selective probe of spin polarization in an altermagnet, and to use spatial conductance line-scans to resolve altermagnetic domain structure in real space.

What's New: Establishes superconductor-tip Andreev reflection as a new experimental paradigm for probing k-dependent spin polarization in altermagnets, and provides the first real-space mapping of altermagnetic domains in CrSb.

Extension Opportunities:

Apply the same MPCS/Andreev reflection methodology to other predicted altermagnet candidates (e.g., RuO2, MnTe, KRu4O8) to systematically build an experimental database of k-dependent spin polarizations across the g-, d-, and i-wave altermagnet families.
Combine this technique with cryogenic scanning probe imaging (e.g., NV-magnetometry or spin-polarized STM) to map altermagnetic domain dynamics under applied strain or electric fields, enabling control of domain switching for spintronic device prototypes.
Engineer CrSb/superconductor heterostructures (e.g., CrSb/Nb thin films) to explore momentum-filtered Cooper pair injection and design Andreev-based spin-polarization detectors or g-wave Josephson junctions.

Replicability: No code/data availability mentioned in the abstract. Reproduction requires high-quality CrSb single crystals oriented along three principal planes, a dilution refrigerator or He-4 cryostat for sub-Tc(Nb ~9.2 K) operation, mechanical point-contact spectroscopy rig with Nb tips, and lock-in based dI/dV measurement electronics — accessible to a well-equipped condensed matter lab but not trivially reproducible.

Research Gaps:

The microscopic mechanism stabilizing the observed ~250-500 nm domain size and ~250 nm domain wall width in a zero-net-magnetization altermagnet remains unexplained.
Quantitative extraction of spin polarization from Andreev reflection relies on BTK-type models that may need extension to fully account for the g-wave momentum structure of altermagnets.

3. Experimental Design Space Exploration of Ultra-Low Threshold Hybrid III-V/Si Quantum Dot Microring Lasers

Authors: Xucheng Yang, Preston Luong, Yatiraj Ramanujam... Published: 2026-06-11 | Citations: 0 arXiv | PDF

Research Question: How can hybrid III-V/Si quantum dot microring lasers be designed to simultaneously achieve ultra-low threshold current, high wall-plug efficiency, strong thermal stability, and high modulation bandwidth for compact 1.3 μm silicon photonic light sources?

Summary: The paper reports hybrid III-V/Si InAs/GaAs quantum dot microring lasers at 1.3 μm achieving record-low thresholds (<0.8 mA, 109 A/cm²), ~10% wall-plug efficiency, and exceptional thermal stability (T₀ = 212 K) through systematic multi-dimensional design exploration. These compact sources also reach 5 GHz modulation bandwidth, making them strong candidates for silicon photonic transmitters.

Key Results: Demonstrated InAs/GaAs quantum dot microring lasers on Si with threshold current <0.8 mA, threshold current density of 109 A/cm², output power >2 mW at 1.3 μm, wall-plug efficiency ~10%, characteristic temperature T₀ = 212 K (record), and 3-dB modulation bandwidth up to 5 GHz.

Key Findings:

Ultra-low threshold currents <0.8 mA with 109 A/cm² density and >2 mW output enable energy-efficient on-chip light sources
Record characteristic temperature T₀ = 212 K indicates the laser threshold is nearly temperature-independent, removing the need for active cooling
High differential gain from the QD active region delivers 3-dB modulation bandwidths up to 5 GHz suitable for direct modulation

Technical Novelty: Multi-dimensional experimental design space exploration (ring geometry, III-V/Si coupling, QD active region) on a hybrid III-V/Si platform yielding simultaneous record values across threshold density, WPE, T₀, and bandwidth — prior work typically optimized one metric at a time.

What's New: First demonstration combining record threshold current density, wall-plug efficiency (~10%), and thermal stability (T₀ = 212 K) in a single hybrid III-V/Si QD microring laser platform via systematic design-space exploration, rather than optimizing a single figure of merit.

Extension Opportunities:

Integrate these microring lasers into a full WDM transmitter PIC with on-chip modulators and multiplexers to validate system-level energy-per-bit metrics
Explore even smaller ring radii and coupled-ring geometries to push toward sub-100 μA thresholds while characterizing RIN and linewidth for coherent applications
Develop CMOS-compatible co-packaged optics modules leveraging the high T₀ to eliminate active cooling in data center interconnects

Replicability: No code/data availability indicated; reproduction requires a III-V-on-Si heterogeneous integration foundry process (MBE-grown InAs/GaAs QDs bonded to SOI), lithography for microring patterning, and RF/optical characterization setups — not feasible without specialized cleanroom access.

Research Gaps:

No reported reliability/aging data or linewidth/RIN characterization needed for coherent and long-lifetime deployments
Limited bandwidth (5 GHz) still below state-of-the-art quantum well DMLs — closing this gap while preserving QD thermal benefits remains open

🔥 GitHub Trending

1. basinlocustmattock/dota2-predictor-analytics

⭐ 6 stars | JavaScript

Machine learning tool for predicting Dota 2 match outcomes, suggesting last picks, mining match data, training models, and visualizing hero statistics such as synergies, counters, and pick patterns.

data-mining dota2 game-analytics logistic-regression machine-learning neural-networks

2. HairdresserKnife/ai-aimbot-detection

⭐ 5 stars | Unknown

Windows computer-vision experiment for Counter-Strike that captures a window, detects people with a YOLO model, selects targets, and controls mouse movement with GPU auto-detection, a live OpenCV GUI,

aimbot computer-vision counter-strike cuda object-detection opencv

3. 97Gz/StockNova

⭐ 4 stars | Python

AI 驱动的 A 股智能投研终端: 17 角色 AI 团队七维并行诊股 + 多空辩论 + 风控会签, 输出可执行评级/仓位/买卖点; 独创回溯诊断用 +5/10/20/60 日真实走势验证 AI 准不准; 44 因子策略 + 真实约束回测 + 7x24 实时快讯; 100% 本地运行, 数据自持, 开源免费

a-shares ai ai-agents backtesting duckdb fastapi

4. seehiong/eduvis

⭐ 3 stars | Python

An open, curriculum-aware schema and reference SVG renderer for educational content.

curriculum education instructional-design learning llm pedagogy

5. mudassar531/hearsay

⭐ 3 stars | Python

crawl4ai for video & audio — one command turns any YouTube video, podcast, or local recording into clean, timestamped, LLM-ready markdown

cli faster-whisper llm markdown mcp podcast

6. zhuofupan/torch-joff-ai

⭐ 2 stars | Python

Spec-first PyTorch experiment toolkit for simulation and process data

experiment-management fault-diagnosis machine-learning mpc process-data pytorch

7. congvmit/awesome-llm-token-reduction

⭐ 2 stars | Unknown

A curated list of techniques, tools, and research for reducing LLM token usage. Optimize context for Claude Code, Copilot, Cursor, and Aider.

ai-coding-assistant awesome awesome-list claude-code context-optimization github-copilot

8. thatrandomasiandev/causal-ml-decision-making

⭐ 1 stars | Python

Causal ML Decision-Making

benchmarks causal-discovery causal-inference machine-learning policy-learning python

9. Ashutosh-789/Robosort-Pro

⭐ 1 stars | HTML

Intelligent object sorting robot using Raspberry Pi, Computer Vision, Machine Learning, Load Cell, Robotic Arm, TFT Display, and Flask Dashboard.

automation computer-vision flask iot machine-learning object-sorting

10. thatrandomasiandev/probabilistic-bayesian-deep-learning

⭐ 1 stars | Python

Probabilistic & Bayesian Deep Learning

bayesian-deep-learning benchmarks calibration conformal-prediction machine-learning python

11. emanuelepiodebernardis/sentineledge-core

⭐ 1 stars | Python

Real-time edge intrusion detection for IoT/IIoT devices

anomaly-detection cybersecurity edge-ai intrusion-detection iot-security kolmogorov-arnold-networks

12. thatrandomasiandev/foundation-model-adaptation

⭐ 1 stars | Python

Foundation Model Adaptation

benchmarks domain-adaptation fine-tuning foundation-models lora machine-learning

13. thatrandomasiandev/graph-ml-relational-systems

⭐ 1 stars | Python

Graph ML & Relational Systems

benchmarks graph-neural-networks link-prediction machine-learning python research

14. thatrandomasiandev/self-supervised-representation-learning

⭐ 1 stars | Python

Self-Supervised Representation Learning

benchmarks contrastive-learning machine-learning python representation-learning research

15. thatrandomasiandev/ml-for-science

⭐ 1 stars | Python

ML for Science

benchmarks climate-modeling genomics machine-learning materials-science protein-structure

Generated by Research Pulse on 2026-06-13 06:07