Science – Page 29

Give Your Photos a Brain: AI Agents Master Image Retouching

21.02.2026 by ebaster

RetouchIQ employs a two-stage training process-supervised fine-tuning guided by user instructions to reason about parameter adjustments, followed by reinforcement learning leveraging a generalist reward model-to optimize a policy for image editing, ensuring both adherence to directives and quality through scalar reward guidance and data filtering.

Researchers have developed an AI system capable of understanding and executing complex photo editing instructions, achieving professional-level results with remarkable consistency.

AI Sheds New Light on Crystal Structures

21.02.2026 by ebaster

A new AI-powered workflow dramatically accelerates the analysis of single-crystal neutron diffraction data, promising faster materials discovery.

Beyond Clicks: Learning User Intent for Smarter Desktop Automation

21.02.2026 by ebaster

IntentCUA decomposes complex, multi-application commands into executable, intent-level plan units by recalling relevant intent units from memory and recomposing them into an end-to-end automation plan, effectively translating high-level goals into sequences of learned skills.

A new framework, IntentCUA, bridges the gap between raw user actions and high-level goals, enabling more robust and efficient automation of complex computer tasks.

Simulating Smarter Policies: Optimizing Agent-Based Models for Environmental Challenges

21.02.2026 by ebaster

The optimization procedure iteratively improved the welfare objective with each sequential iteration, demonstrably surpassing the performance of a scenario lacking any policy intervention, as indicated by the dotted baseline.

A new framework leverages Bayesian optimization to efficiently refine policies within complex simulations, offering a powerful tool for addressing real-world environmental issues.

Mastering Tools Without Training: A New Approach to Robotic Dexterity

21.02.2026 by ebaster

A goal-conditioned reinforcement learning policy is trained in a simulated environment with procedurally-generated objects and random goal poses, then deployed without further training to manipulate real-world tools from the DexToolBench suite by replicating trajectories observed in human videos.

Researchers have developed a novel framework that enables robots to manipulate unfamiliar tools and complete tasks without task-specific training data.

Beyond Prediction: Reframing Time Series with Dynamical Systems

21.02.2026 by ebaster

Long-term temperature predictions were generated using both Dynamic System Regression (DSR) and Time Series (TS) models, demonstrating the capacity of each approach to forecast weather patterns over extended periods.

A new perspective on time series analysis argues that understanding the underlying system dynamics, rather than simply forecasting future values, is crucial for robust and generalizable models.

From Steel to Softness: Bridging the Grasp Gap for Robots

21.02.2026 by ebaster

$The Conditional Flow Matching (CFM) framework synthesizes grasps by learning a continuous transformation-guided by a conditional velocity field [latex]\mathbf{v}_{\theta}[/latex]-from initial rigid poses generated by Anygrasp [latex]\mathcal{G}_{\text{Anygrasp}}[/latex] to target soft gripper poses [latex]\mathcal{G}_{\text{CFM}}[/latex], a process parameterized by a feed-forward MLP and conditioned on scene geometry encoded by a U-Net autoencoder into a latent vector.$

New research demonstrates a method for transferring grasp planning from traditional rigid robot grippers to the more versatile, but challenging, domain of soft robotics.

Ask and Verify: Smarter SQL Queries with Agentic Data Exploration

21.02.2026 by ebaster

Existing text-to-SQL paradigms-static schema prompting, which suffers from hallucinations due to ambiguous metadata, and pre-processed schema augmentation, which introduces extraneous noise-are superseded by a new approach, APEX-SQL, that leverages agentic exploration and a hypothesis-verification loop to actively interrogate the database and establish a grounded foundation for logical reasoning.

A new framework, APEX-SQL, improves the accuracy of converting natural language questions into database queries by actively learning and validating its assumptions about the underlying data.

Robots Learn by Watching: Building Task Models from Limited Examples

21.02.2026 by ebaster

The system extracts task skeletons from demonstrated trajectories by constructing manipulation graphs and identifying event sequences tied to changes within those graphs; objects are matched across demonstrations using pre-trained features, enabling the grouping of related events and ultimately the inference of grasping and placement actions with associated target poses.

A new approach enables robots to infer complex tasks by observing just a few demonstrations, focusing on how objects relate and change during execution.

Symmetry’s Guiding Hand: AI Predicts Stable Materials with Unprecedented Accuracy

21.02.2026 by ebaster

A generative framework, guided by principles of symmetry, forms the basis of the proposed approach.

A new generative AI framework harnesses the power of crystal symmetry to dramatically improve the prediction of stable and novel materials.