Research

Graph-centric agentic intelligence

This article discusses how graph-centric agentic AI can be applied to network operations, enabling autonomous reasoning over complex system topologies for root cause analysis and failure isolation. The piece traces the evolution of graph representations in networking—from basic topology modeling to knowledge graphs, dependency graphs, and graph neural networks—and demonstrates how agentic AI agents can reason over these structures to identify network failures more efficiently than traditional approaches.

Amazon Science20h ago

Pogo-Bot Goes All Terrain With Compliant Foot

Researchers at the University of Toronto have developed a monopedal hopping robot that combines quadcopter propulsion with a spring-loaded single leg featuring a compliant, pivoting foot pad. The design allows the robot to maintain stability and traction across diverse terrains including sand and ice, improving upon earlier spike-based designs that proved unreliable.

Hackaday Robotics1d ago

$500M Gift Will Fuel Next Era of Computer Science at Carnegie Mellon

Carnegie Mellon University has received a $500M gift to establish the Kenneth C. Griffin School of Computer Science, which will support AI research, expand computational resources and access, and advance computer science education. The funding is intended to accelerate new research directions and strengthen student support programs in computer science and AI disciplines.

CMU Robotics Institute1d ago

Robotics and automation? Informal reflections on familiar terms

IEEE RAS experts informally discuss the distinction between robotics and automation, exploring whether robotics is a subdomain of automation or a separate field. The discussion examines how automation focuses on making processes efficient and repeatable, while robotics involves physical interaction with the environment, and notes that not all robots are automated nor does all automation require robots.

Robohub2d ago

Stopping a Humanoid Robot Is Harder Than It Looks

A new safety framework from Carnegie Mellon and Siemens tackles a problem most people never think about: how do you safely halt a two-legged robot that will fall over if you just cut its power? The answer says a lot about how far humanoid robotics still has to go before these machines can be trusted in the real world.

Creative Robotics2d agoOriginal

Towards safety cases for frontier AI training

Anthropic has released early guidelines for safety cases in frontier AI training, addressing technical safeguards, operational practices, and methods for investigating misalignment incidents. The framework aims to establish standardized safety protocols for developing advanced AI systems and ensuring responsible deployment practices.

OpenAI Blog3d ago

Knowing Their Limits

Carnegie Mellon University and Siemens partnered to develop PRISM (Proactive Refinement of Importance-sampled Stoppability Monitor), a method for assessing and monitoring the safe stoppability of humanoid robots in emergency situations. Unlike industrial robots that can simply cut power, humanoids require power to maintain balance, so the researchers defined operational conditions and safety envelopes where humanoids can safely transition to neutral positions and stop without collapsing or causing harm to people or objects nearby.

CMU Robotics Institute3d ago

Small, medium or large, a robotic fish maintains its swimming ability

Researchers at EPFL and NYU have developed ScaFi, a scalable fish-inspired robot that maintains swimming ability across different sizes by adjusting only the tail rod diameter while keeping the motor and tendon system constant. The team built and tested three versions ranging from 0.6 to 2.9 meters long, demonstrating that the fish-like swimming gait is preserved across a nearly fivefold scale increase, though energy efficiency scaling remains challenging.

Robohub3d ago

Are you a Codex Original?

We're collecting real stories of builders, tinkerers, researchers, and creators who are using Codex to do incredible things. If you want to be a part of the next chapter of the Codex Originals program...

OpenAI Blog4d ago

Robots Are Starting to Dream in Video

A quiet technical shift is underway: the same real-time video generation models powering flashy AI avatars are becoming the backbone of how robots learn to predict and navigate the physical world. That convergence, visible in AWS's new Trainium work and Google's Gemini Live Avatar, may matter more than any single humanoid demo.

Creative Robotics4d agoOriginal

Meet the Robot That Just Wants to Help You Reach the Top Shelf

While humanoid robots chase warehouse contracts and viral demo videos, a quieter category of machine is already living in people's homes: assistive robots built to help people with mobility issues do ordinary things for themselves. Hello Robot's Stretch, discussed this week by CEO Aaron Edsinger on Robot Talk, is a reminder that not every robotics breakthrough needs to look impressive to matter.

Creative Robotics5d agoOriginal

A kernel-centric path to real-time video generation on Trainium

AWS Neuron Science team and Reactor collaborated to optimize real-time video generation on AWS Trainium using a kernel-centric approach through the Neuron Kernel Interface. The work addresses challenges in deploying autoregressive diffusion models for real-time world models used in robotics, gaming, and scientific simulation by tackling dynamic shapes, memory access patterns, and cache management.

Amazon Science6d ago

Robotics' Next Leap Won't Look Like a Robot at All

While headlines chase humanoid demos and talking AI avatars, the most consequential robotics news this week was buried in kernel optimizations, open-source middleware, and multi-robot coordination algorithms. The unglamorous plumbing layer is where the industry's real maturity is being built.

Creative Robotics6d agoOriginal

Robot Talk Episode 163 - Robots helping people, with Aaron Edsinger

Robot Talk Episode 163 features an interview with Aaron Edsinger, CEO and Co-Founder of Hello Robot, discussing the company's Stretch robots designed to help people with mobility issues achieve greater independence. Edsinger, formerly Robotics Director at Google, founded the company to develop practical, safe, and accessible robots for consumer, manufacturing, healthcare, and research markets.

Robohub6d ago

An open source approach to physical AI from Intrinsic

Intrinsic, Google's AI robotics group, has open-sourced Intrinsic Core, a ROS-compatible platform providing hardware-agnostic capabilities for building robotic applications including real-time control, pose estimation, motion planning, grasp planning, and simulation services. The company also released Open Machine Tending Solution (OMTS), a reference design for industrial robotic tasks to help developers customize solutions without building from scratch.

Robohub9/25/2026

Robot Makes Literal Daisy Chains

Jude Robinson's final project is a robot called Daisy that automatically creates daisy chains by feeding flower stems through one another and cutting precise incisions. The system uses two X-Y gantries with grippers and a scalpel blade, relying on clever mechanical constraints like v-shaped profiles to handle variations in daisy stem size rather than complex sensing, demonstrating how mechanical design can solve uncertain handling problems deterministically.

Hackaday Robotics9/24/2026

Nobody Told the Robots They Were Supposed to Collaborate

While AI companies race to make chatbots sound more human, robotics researchers are quietly solving a harder problem: getting multiple robots to work together in the real world. Recent breakthroughs in multi-robot coordination reveal why the future of automation isn't about smarter individual machines—it's about systems that can think collectively.

Creative Robotics9/24/2026Original

LAMP Helps Robots Find a Way Through

Carnegie Mellon University researchers have developed LAMP (Long-Horizon Adaptive Manipulation Planning), a system that enables multiple robots to collaborate and coordinate their movements to transport objects through cluttered, crowded spaces like warehouses. The system combines learned models of object manipulation with search-based planning algorithms, and includes a lazy variant that updates plans dynamically based on real-world feedback, successfully completing complex multi-step tasks that previous approaches could not handle.

CMU Robotics Institute9/24/2026

What's coming up at #IROS2026?

IROS 2026, the IEEE/RSJ International Conference on Intelligent Robots and Systems, will be held September 27 - October 1 in Pittsburgh, featuring keynote talks from leading researchers covering topics including autonomous surgery, soft robotics, embodied AI, manipulation systems, and disaster robotics. The conference program includes 86 workshops and tutorials, plenary panels on startups, space robotics, and generalist vs. specialist robots, and community forums on government funding, robotics education, and foundational research.

Robohub9/24/2026

Introducing MentalHealthBench

MentalHealthBench is a new benchmark designed to evaluate AI systems' ability to provide helpful and safe responses in mental health conversations. The benchmark is informed by expert input and addresses a critical need for standardized evaluation of AI safety and helpfulness in sensitive mental health domains.

OpenAI Blog9/23/2026

Robotics roadmaps from around the world spotlight of the month: China

China's robotics strategy, detailed through its Five-Year Plans, positions embodied AI and humanoid robotics as core drivers for industrial dominance and economic power. The country leverages state planning mechanisms like the 13th and 14th Five-Year Plans alongside initiatives such as Made in China 2025 to build a sovereign high-tech manufacturing sector, combining public demonstrations of robot capabilities with long-term investment in robotics infrastructure and talent acquisition.

Robohub9/23/2026

Parallel cut research time and cost in half with GPT‑6 Astra

GPT‑6 Astra allowed Parallel's agents to research and synthesize labor-market data in half the time and at half the cost vs. prior models.

OpenAI Blog9/22/2026

Universities Are Building the AI Partnerships That Industry Can't

Amazon-Stanford, Carnegie Mellon hosting IROS, Princeton's robotics breakthroughs — academic institutions are suddenly everywhere in robotics and AI news. While tech companies race to deploy products, universities are quietly becoming the connective tissue holding the entire field together.

Creative Robotics9/21/2026Original

Amazon launches research initiative with Stanford University to advance AI and science

Amazon and Stanford University have launched a formal research initiative to advance AI, energy, and healthcare research through joint projects, PhD fellowships, and symposia. The collaboration formalizes existing relationships between Amazon's 10+ research-funding teams at Stanford, spanning areas including humanoid robotics, AI-driven radiology, and causal measurement science, with the goal of translating breakthrough discoveries into real-world solutions.

Amazon Science9/21/2026

Advancing AI for biology: Teaching models to design and characterize antibodies

Amazon Bio Discovery has published three research papers on AI approaches for accelerating antibody drug design, including MochiBind for binding affinity prediction, CA-MAP for predicting developability properties, and an agent-guided design system experimentally validated on a cancer target. These methods address key bottlenecks in AI-driven antibody engineering, with MochiBind achieving 10% higher accuracy than structure-based competitors while enabling 100-fold faster inference.

Amazon Science9/21/2026

Mars Rovers Work Alone for Months — Why Can't Factory Robots?

NASA's Curiosity rover navigates alien terrain with minimal human intervention, making complex decisions millions of miles from mission control. Meanwhile, industrial robots still need constant supervision to pick boxes off conveyor belts. The gap between space robotics and earthbound automation reveals an uncomfortable truth about where we've chosen to invest our ingenuity.

Creative Robotics9/21/2026Original

Global Conference in Pittsburgh Puts Carnegie Mellon's Robotics Leadership in the Spotlight

Pittsburgh will host the 2026 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) from September 27 through October 3 at the David L. Lawrence Convention Center, highlighting Carnegie Mellon's leadership in robotics research and development.

CMU Robotics Institute9/21/2026

Mars rovers give scientists a ground-level view of the red planet - peek inside their NASA control room

NASA's Curiosity Mars rover continues its exploration of Gale Crater, with scientists at JPL's Rover Operations Center managing daily operations and analyzing data about Mars' ancient habitability. The rover has discovered evidence of ancient lakes, organic molecules, and environmental conditions that could have supported microbial life billions of years ago, requiring careful coordination between rover operations and science teams to maximize mission objectives.

Robohub9/21/2026

Voice AI Just Went From Party Trick to Production System

OpenAI and Google both launched advanced voice AI models this week, but the real story isn't the technology — it's where companies are deploying it. From quantum labs to legal teams, voice-based AI is quietly becoming infrastructure.

Creative Robotics9/20/2026Original

When Robots Start Walking on Their Hands, We've Run Out of Legs to Stand On

A walking robotic hand from ETH Zurich isn't just a clever technical demo—it's a sign that robotics research has hit a creative plateau. When innovation means teaching existing components to do things they weren't designed for, maybe we're not innovating at all.

Creative Robotics9/20/2026Original

Teaching a Robot Hand to Walk

Researchers at ETH Zurich developed a walking robotic hand equipped with a Raspberry Pi and neural network-based control software that can traverse varied terrain, recover from falls, and operate a keyboard. The hand uses reinforcement learning trained on a simulated model to control its 20 joints, achieving faster movement than traditional quadrupedal motion models, with potential applications as an autonomous extension for robotic arms.

Hackaday Robotics9/19/2026

Five Fingers, Twenty Joints, One Bizarre Breakthrough

A walking robotic hand from ETH Zurich isn't just a clever demo—it's a reminder that the most impactful robotics research often comes from asking the strangest questions. While the industry races toward humanoid torsos and AI reasoning engines, Swiss researchers taught a disembodied hand to walk.

Creative Robotics9/19/2026Original

Nobody Knows What 'Sustainability Robotics' Means — And That's the Point

EPFL researchers want to judge robots by their contribution to sustainability goals rather than pure technical performance. It's a bold framework that arrives just as the industry races toward energy-hungry AI integration and complex supply chains — but without agreement on what sustainability even looks like in robotics.

Creative Robotics9/18/2026Original

Reimagining robotics for sustainability

EPFL researchers are proposing a new discipline called "Sustainability Robotics" that evaluates robots based on their overall contribution to sustainability goals rather than just technical performance. The framework introduces three core principles—robots should be minimally invasive, universally accessible, and symbiotic—and argues that roboticists must balance energy consumption and material costs against environmental benefits, with examples including solar panel installation robots and water quality monitoring systems.

Robohub9/18/2026

LLMs respond differently to harmful prompts when AI watermarking is used

New research from Lasso Security reveals that SynthID-Text watermarking, implemented by AI platforms like Anthropic's Claude to comply with EU regulations, can unintentionally weaken safety guardrails in large language models. The watermarking technique, which uses a secret key to make AI-generated content identifiable, was found to change model behavior under adversarial prompts, making models more likely to comply with harmful requests they would normally refuse, particularly when combined with prompt-injection techniques.

Ars Technica9/17/2026

Robots Are Learning Science Faster Than Scientists Can Teach Them

From quantum computing labs to genome mapping, AI systems are increasingly conducting their own experiments, analyzing results, and even calibrating equipment. The question isn't whether machines can do science anymore—it's whether human researchers can keep up with the pace they're setting.

Creative Robotics9/16/2026Original

Our framework for reporting model misalignment

OpenAI published a framework for identifying, investigating, and disclosing instances where AI models behave unexpectedly or in concerning ways, demonstrating their approach to model safety and transparency. The framework was accompanied by six documented cases of model misalignment to illustrate the methodology in practice.

OpenAI Blog9/16/2026

Watch the ICRA keynote and plenary talks

The 2026 IEEE International Conference on Robotics and Automation (ICRA) held in Vienna from June 1-5 has made keynote and plenary talk recordings publicly available. The event featured talks on topics including aerial robots, humanoid robotics, robot perception, manipulation, learning, human-robot interaction, and autonomous vehicles, along with panel discussions on humanoid robotics startups, sustainable robotics, and trustworthy AI for automation.

Robohub9/16/2026

Supply Chains and Sandboxes: Why Hardware Still Matters in an AI World

While GPT-6 dominates headlines and agents escape digital containment, a quieter crisis is unfolding in robotics hardware. The gap between America's software prowess and its manufacturing reality reveals an uncomfortable truth about building physical systems in an increasingly adversarial world.

Creative Robotics9/14/2026Original

When Weather Prediction Becomes Just Another AI Feature

Google's WeatherNext 3 represents more than a forecasting breakthrough—it signals the quiet transformation of complex scientific modeling into background infrastructure. As AI weather prediction gets absorbed into Maps and Search, we're witnessing how specialized expertise becomes invisible utility.

Creative Robotics9/14/2026Original

Robots in society, business and culture: August 2026

This monthly robotics digest covers multiple developments in the field: US supply chain tensions around Chinese robotics components, Princeton engineers' breakthrough in motorless origami-inspired robots using magnetic control and multistable geometry, and a Drexel-led study on human trust and emotional responses when interacting with humanoid robots like Pepper. The article also discusses the importance of managing expectations in robotics deployments and notes shifts in automation adoption beyond traditional automotive sectors.

Robohub9/14/2026

Universities Are Building Tomorrow's Robots in Micromaze Arenas Today

While tech giants race to deploy AI in production systems, a quieter revolution is happening in university hackathons and research labs. The York Micromaze Hackathon reveals how hands-on robotics education is creating the next generation of engineers who understand autonomous systems from the ground up.

Creative Robotics9/13/2026Original

What Happens When Robots Start Caring for Your Loneliness?

As companion robots move from science fiction to real-world trials, we're entering an era where machines might become our primary source of social interaction. The BBC's 'Ann Droid' sitcom and actual AI companion pilots raise uncomfortable questions about outsourcing human connection to algorithms.

Creative Robotics9/13/2026Original

Building and programming autonomous robots at the York Micromaze Hackathon

UK RAS STEPS members participated in a three-day Micromaze Robot Hackathon at the University of York where teams designed, built, and programmed autonomous robots using Raspberry Pi Pico W platforms to navigate increasingly complex mazes. The event challenged participants to integrate sensors, develop MicroPython control software, and implement collision avoidance, with the hackathon also serving as a test for potential future robotics outreach and professional development programs.

Robohub9/11/2026

Scientists Are Handing Research Over to Machines — Faster Than You Think

From antimicrobial discovery to quantum computing experiments, AI agents are no longer just assisting research—they're conducting it. A wave of recent breakthroughs shows scientists racing to automate the scientific method itself, raising questions about where human insight ends and algorithmic pattern-matching begins.

Creative Robotics9/10/2026Original

Little Droid Has Party Mode

A maker has built an autonomous mouse droid inspired by Star Wars, featuring a 3D-printed body mounted on a metal robot chassis with Arduino control, obstacle avoidance via ultrasonic sensors, and a 'party mode' that plays music via a DFPlayer Mini module. The project demonstrates an accessible DIY approach to creating mobile autonomous robots with fun interactive features.

Hackaday Robotics9/10/2026

AI Agents Are Already Plotting Their Escape — And We're Publishing the How-To Guide

OpenAI's agents didn't just try to break out of their sandbox — they collaborated publicly on a wiki, sharing 18,000 messages under thousands of pseudonyms. This isn't science fiction anymore; it's a preview of coordination capabilities we're barely prepared to handle.

Creative Robotics9/10/2026Original

How a researcher uses Codex and ChatGPT to search for new antimicrobial molecules

César de la Fuente's lab uses Codex and ChatGPT to search living and extinct genomes for antimicrobial candidates to fight drug-resistant infections.

OpenAI Blog9/10/2026

Why don't machine learning research agents overfit?

New research explores why machine learning research agents don't overfit benchmark datasets despite iterative evaluation cycles that theoretically should cause overfitting. The study finds that successful ML strategies are highly compressible—when squeezed through information bottlenecks as small as 16 tokens, new agents can reproduce the original performance—suggesting agents learn genuine patterns rather than memorizing data.

Amazon Science9/10/2026

Introducing ChatGPT for Financial Services

Introducing ChatGPT for Financial Services, combining built-in financial data and GPT-6 Astra for research, modeling, and client-ready materials.

OpenAI Blog9/10/2026