Smell Check: AI Designs Novel Odorant Molecules
![Generated molecules exhibit a nuanced relationship between structural distance from seed compounds and predicted odor probabilities, as evidenced by a UMAP visualization of the latent space and a statistically significant correlation-demonstrated through [latex]\sigma\_{\phi}(z|X)[/latex] distance metrics and QSAR modeling-between generated compounds, existing ChemBL molecules, and known odorants from the Good Scents database.](https://arxiv.org/html/2512.23080v1/Fig_5.png)
Researchers have developed an artificial intelligence framework to generate new odorant molecules with predicted desirable properties, tackling the challenges of limited data in olfactory research.
Science
Building Better Teammates: AI Learns to Collaborate On the Fly
New research demonstrates how artificial intelligence can dynamically adapt to different collaborators in complex team environments, achieving strong performance through learned teammate modeling.
Can You Spot the Fake? Humans Fail to Detect AI-Generated Images
A new study reveals that people are surprisingly poor at distinguishing between authentic photographs and those created by artificial intelligence.
Standardizing Robot Descriptions for a More Connected Future
![The Universal Robot Description Directory (URDD) structures robot information into modular, version-controlled subdirectories utilizing JSON/YAML formats, encompassing kinematic chains, joint limits, and preprocessed geometric data like [latex]convex hulls[/latex], to facilitate reusable data for robot planning, control, and visualization while eliminating redundant derivations from raw URDF files.](https://arxiv.org/html/2512.23135v1/x1.png)
A new directory aims to streamline robot modeling and data exchange, fostering greater collaboration and accelerating robotics development.
Building Worlds with Words: The Rise of Web-Based AI Environments
Researchers are exploring how to combine the power of large language models with web technologies to create dynamic and controllable virtual worlds for artificial intelligence agents.
Seamless Rides: Balancing Human Control with Autonomous Driving
A new approach to cooperative driving leverages machine learning to anticipate driver needs and dynamically share control, creating a safer and more comfortable experience.
Life’s Blueprint: How Networks Shape Living Systems
![Living systems exhibit complex dynamics governed by constraints at multiple spatial scales, where each level [latex]\mathcal{S}_{i}[/latex] integrates novel limitations [latex]\Omega(\mathcal{S}_{i})[/latex] with those inherited from lower organizational levels [latex]\Gamma(\mathcal{S}_{i})[/latex], and is potentially described by sets of non-autonomous, stochastic differential equations, a framework that multilayer network modeling seeks to capture through the analysis of interdependencies across these hierarchical layers.](https://arxiv.org/html/2512.22651v1/x5.png)
A new perspective reveals that the architecture of biological networks-governed by energy, information, and evolution-underpins the robustness and adaptability of all living organisms.
Beyond Reward: How ‘Emotions’ Can Build More Resilient AI
![Following a disruptive event-simulated here as a gravity shift-the agent demonstrates robust recovery through a dynamic interplay of internal stress accumulation [latex]\sigma_{t}[/latex], a near-instantaneous plasticity response that amplifies the learning rate by 600%, and subsequent total reward stabilization.](https://arxiv.org/html/2512.22200v1/figure1_recovery.png)
A new framework integrates principles of biological homeostasis into reinforcement learning, allowing agents to adapt and thrive in ever-changing environments.
Simulating Materials, Simplified: An AI Agent Takes Control
A new AI-powered agent is automating complex materials simulations, promising faster discovery and improved research reproducibility.
Swarm Intelligence: Scaling Robot Teams for Optimal Performance
![The algorithm optimizes the allocation of [latex] N=24 [/latex] agents across [latex] T=4 [/latex] tasks-each defined by a scalability curve [latex] C(d_i, n_i) [/latex]-favoring simpler tasks when resources are limited, ultimately achieving an optimal distribution of [latex] N^* = [11, 9, 3, 1] [/latex], but shifting towards more challenging tasks as the swarm size increases and resources become abundant.](https://arxiv.org/html/2512.23431v1/x1.png)
A new algorithm efficiently allocates tasks to robot swarms, ensuring performance scales effectively as team size grows.