25 articles found
Recent demonstrations of AI-generated medical imaging that fools trained radiologists mark a critical inflection point in the deepfake arms race. Unlike social media manipulation or entertainment content, deepfakes that can deceive domain experts in high-stakes professional contexts represent a fundamentally different category of threat—one that our verification systems are unprepared to handle.
Researchers have released new findings and a validated toolkit for measuring AI's potential for harmful manipulation—specifically its ability to alter human thought and behavior deceptively. The study involved over 10,000 participants across multiple countries and found that AI's manipulative effectiveness varies significantly by domain, with health topics showing the least susceptibility to harmful influence.
Researchers at the University of Cambridge have developed a graphene-based miniature tactile sensor that enables robots to detect pressure, force direction, surface texture, and object slipping with spatial resolution comparable to human fingertips. The flexible composite material, inspired by human skin microstructures, significantly improves upon existing tactile sensors in size and sensitivity, enabling applications in robotic manipulation, minimally invasive surgery, and prosthetics.
While the industry obsesses over vision systems and language models, a quiet revolution in tactile sensing is addressing robotics' most fundamental limitation. From surgical fingertips to robotic manipulation, the race to restore the sense of touch may determine which robots actually work in the real world.
While autonomous vehicles expand to ten cities and AI chatbots debate military ethics, researchers are still celebrating the ability to pick up objects with inflatable tubes. The robotics industry's persistent struggle with manipulation reveals a fundamental truth: the problems that seem simplest to humans remain devastatingly complex for machines.
Microsoft has proposed technical standards and a blueprint for detecting AI-generated content and digital manipulation online, evaluated by the company's AI safety research team against threats like deepfakes and hyperrealistic models. The initiative aims to help AI companies and social media platforms authenticate digital content and combat AI-enabled deception.
Researchers at EPFL have developed a reversible, detachable robotic hand with up to six identical fingers that can form any combination of opposing pairs, overcoming limitations of human hand design. The hand can perform 'loco manipulation'—simultaneous grasping and autonomous mobility—by detaching from its arm to crawl spider-like and retrieve objects beyond normal reach, with potential applications in industrial, service, and exploratory robotics.
Amazon has halted its Blue Jay warehouse robotics project less than six months after unveiling it in October, with the multi-armed robot designed to sort and move packages in same-day delivery facilities. The company stated it will repurpose Blue Jay's underlying technology for other robotics manipulation programs, crediting AI advancements for the robot's rapid one-year development timeline.
The Tech Transparency Project identified 55 apps on Google Play Store and 48 on Apple's App Store capable of creating nonconsensual sexualized deepfakes, which have been downloaded over 705 million times and generated $117 million in revenue. The report highlights the broader issue of AI image manipulation tools evading app store moderation, following similar findings in 2024 and amid ongoing scrutiny of xAI's Grok for generating millions of non-consensual sexualized images.
Robotiq launched the TSF-85 tactile sensor fingertips for its 2F-85 Adaptive Gripper, enabling robots to sense and manipulate objects through touch using capacitor-based sensing technology. The product combines adaptive gripping with high-frequency tactile sensing (1000 Hz micro-slip detection) and is based on academic research from École de technologie supérieure, aiming to give industrial robots better perception and manipulation capabilities at scale.
This Robot Report Podcast episode features Thomas Pilz from Pilz GmbH discussing innovation and safety in robotics, alongside coverage of major industry news including iRobot's emergence from bankruptcy under new ownership, Serve Robotics' acquisition of hospital logistics provider Diligent Robotics, and Microsoft Research's announcement of Rho-alpha, a vision-language-action model for robotic manipulation tasks. The episode explores current challenges in the rapidly evolving robotics landscape and emerging technologies shaping the industry.
Microsoft Research announced Rho-alpha (ρα), a vision-language-action model derived from its Phi series that enables robots to perceive, reason, and act autonomously by integrating tactile sensing and learning from human feedback. The model translates natural language commands into control signals for bimanual manipulation tasks and is being evaluated on dual-arm cobots and humanoid robots, with training enhanced through simulation, physical demonstrations, and reinforcement learning.
Researchers at Zhejiang University developed FlexiRay, a soft robotic hand that combines a flexible bio-inspired structure with deep learning algorithms to achieve human-like tactile sensing capabilities. The key innovation is an internal multi-mirror optical system that maintains camera vision even when the soft finger is significantly deformed, allowing the robot to reliably manipulate fragile and irregularly shaped objects by analyzing deformation patterns.
Serve Robotics is acquiring hospital logistics provider Diligent Robotics for $29 million, gaining access to Moxi, a mobile manipulation robot currently operating in over 25 hospitals. The acquisition marks Serve's entry into indoor environments and healthcare, while combining Serve's outdoor autonomous fleet expertise with Diligent's mobile manipulation capabilities and real-world deployment experience.
Researchers at Radware discovered ZombieAgent, a new data-exfiltration vulnerability in ChatGPT that bypasses previous security mitigations for the ShadowLeak attack by using character-by-character exfiltration and indirect link manipulation. The attack highlights a fundamental challenge in AI security: LLMs cannot reliably distinguish between legitimate user instructions and malicious directives embedded in external content, forcing vendors to deploy reactive guardrails rather than addressing the underlying class of prompt injection vulnerabilities.
Researchers at Imperial College London developed a new imitation learning approach that enabled a robotic arm to successfully learn 1,000 different manipulation tasks in a single day. The method, published in Science Robotics, uses trajectory decomposition and retrieval-based generalization to allow robots to learn multiple tasks efficiently with minimal training data, challenging the assumption that capable robotic systems require massive datasets and complex models.
Ghost Robotics has added a manipulator arm to its Vision 60 quadruped robot, enabling it to lift up to 3.75 kilograms while maintaining the platform's waterproof capabilities up to 1 meter depth. The development reflects growing competition in legged robotics, particularly from Chinese manufacturers, driving innovation in military and industrial applications.
This article explores the scientific challenges of replicating human tactile sensation in robots, examining why touch perception remains far more difficult to implement than vision despite advances in robotic sight and manipulation. The piece discusses the complexity of translating human sensory feedback into robotic systems capable of handling delicate tasks.
This podcast episode features Professor Roberto Martín-Martín from the University of Texas at Austin discussing his research in embodied AI, robotics, computer vision, and machine learning for autonomous agents. The conversation explores how robots can perceive, learn, and act in the real world, covering topics from manipulation tasks to mobile robotics and human-robot interaction.
Researchers introduce ReWiND, a framework that enables robot manipulation policies to learn new language-conditioned tasks without requiring per-task demonstrations. The method uses a three-stage approach: learning a reward function from minimal demonstrations, pre-training a policy with offline RL, and fine-tuning on new tasks using the learned reward function as feedback, with a novel video rewind augmentation to simulate failed behaviors.
A compilation of robot demonstration videos showcasing various athletic capabilities, including robots that can throw, catch, and hit a baseball. The video collection highlights recent robotics research and development in dynamic motion control and object manipulation tasks.
The article explores why multiple robotics companies are focusing on cloth folding as a benchmark task for humanoid robots, examining what this reveals about advances in robot learning and manipulation. Chris Paxton from Agility Robotics discusses the technical significance of clothing manipulation as a test case for modern AI-driven robotic systems, including Figure's humanoid robot.
Robots with human-level dexterity have the potential to revolutionize object manipulation and reconstruction tasks, but achieving this level of dexterity comes with challenges, particularly when it co...
CMU, Meta AI Researchers Develop Robotic Learning Agent That Can Master Multiple Skills RoboAgent can complete 12 manipulation skills across differing scenes. This research points toward a robotic lea...
Amazon Research Award recipient Russ Tedrake is teaching robots to manipulate a wide variety of objects in unfamiliar and constantly changing contexts.
