50 articles found
Amazon announces 34 Build on Trainium recipients from 30 universities including Stanford, MIT, and UC Berkeley, awarding $110 million in compute credits to advance responsible AI research. The initiative focuses on critical areas including AI safety, multilingual models, sustainability, synthetic data generation, and large-scale distributed systems on AWS Trainium infrastructure.
Zoox, an Amazon subsidiary, unveiled the final production design of its purpose-built robotaxi after gathering feedback from over 500,000 riders. The company plans to scale production to 100 vehicles per week and expand autonomous taxi service from Las Vegas to Austin, Miami, Dallas, and Phoenix, while coordinating with NHTSA for full regulatory approval.
Researchers at Amazon have developed ControlG, a framework that applies industrial control system principles (PID controllers) to coordinate multiple conflicting machine learning objectives in graph neural networks. Rather than blending gradients from different training goals at each step, ControlG allocates computational capacity sequentially and dynamically, eliminating common multitask learning failures like negative transfer and objective starvation.
OpenAI is handing out free advanced AI access to 100,000 academics while Amazon backs formal verification research. It looks like generosity, but the real story is about who controls the future of scientific discovery.
Amazon is providing substantial long-term financial support to the Lean Focused Research Organization to advance Lean, a programming language that enables mathematical proofs of software correctness for AI systems. Amazon is using Lean-based verification across multiple products including Bedrock AgentCore for agentic safety, SampCert for differential-privacy guarantees, and AWS Neuron for AI chip compilation, with the language increasingly important for ensuring trustworthy and verified AI agents.
While the robotics industry has obsessed over vision and language models, touch has remained the forgotten sense. New research from Amazon and MIT suggests that's about to change — and it might be the breakthrough that actually matters for getting robots into the real world.
From browser-based quadruped simulators to Amazon's tactile physics engines, the robotics industry is rapidly recognizing that the real innovation happens before hardware ever touches the real world. The shift toward sophisticated simulation environments represents a fundamental change in how robots learn—and who gets to teach them.
Amazon and University of Michigan researchers developed HydroShear, a physics-based simulator that teaches robots to sense and understand tactile forces for complex manipulation tasks. The simulator enables robots to learn dexterous manipulation policies entirely in simulation and transfer them to real robots with 93% success rate, addressing a major challenge in contact-rich robotic tasks.
Amazon is reportedly developing a more powerful version of Alexa called 'Moonraker' that would enhance the AI assistant's agentic capabilities to handle multi-step, complex tasks like booking rides and sending messages simultaneously. The project aims to help Alexa compete with agentic AI models from Google, Anthropic, and OpenAI, though internal documents suggest Amazon may need to delay or scale back the initiative due to projected GPU costs exceeding $100 million.
Meta's reported entry into cloud computing isn't just about infrastructure—it's a signal that the entire AI economy is reorganizing around whoever controls the servers. As companies from Meta to Amazon build competing cloud empires, the question isn't who builds the best AI model, but who can afford to run it.
There are now enough Amazon Leo satellites to launch the service, but its constellation is still tiny compared to Starlink's.
Meta is reportedly building its own cloud business to leverage infrastructure investments made for AI model training and deployment, potentially offering services like AI model access and data center compute leasing to compete with Amazon Web Services, Google Cloud, and SpaceX. The initiative is being developed under Meta Compute, Meta's data center and AI-focused division created in January, and aligns with the company's $600 billion US investment commitment through 2028.
German auto supplier Bosch is pivoting toward humanoid robotics and related sensor technology as its traditional auto parts business faces pressure from competition and weak demand. The company plans to expand production of MEMS sensors crucial for robotic dexterity and has partnered with robotics firm Neura, which recently raised $1.4 billion in funding from backers including Bosch, Nvidia, Amazon, and Tether to accelerate deployment of robots and physical AI systems.
Rocket Lab will own satellites connecting the most remote corners of the Earth.
Amazon's Zoox has unveiled an updated version of its autonomous robotaxi designed for large-scale production, featuring improved comfort, better visibility through rotating color-changing reflectors, and enhanced interior ergonomics. The company plans to begin manufacturing the new model at its Hayward, California facility and can scale production to 100 vehicles per week to support expansion in 2026.
NEURA Robotics announced a Series C funding round of up to $1.4 billion to accelerate development of physical AI and cognitive robots. The company is building an ecosystem called the Neuraverse that combines robotics, AI, sensors, and edge computing for manufacturing, logistics, and service applications, with strategic partnerships including Qualcomm, Amazon, NVIDIA, and Bosch.
Luca Guadagnino's Sam Altman biopic has to find a new studio after Amazon dropped it.
Amazon researchers introduce Simple Strands Agent (SSA), a customizable harness designed to minimize the intent-execution gap in agentic AI systems, where mismatches occur between what language models intend and what the software harness actually executes. The research demonstrates that improving tool interfaces, feedback mechanisms, and balancing internal reasoning with external interactions can achieve state-of-the-art performance across multiple benchmarks, while also showing that effective agent design requires model-specific customization rather than one-size-fits-all approaches.
Amazon's natural-language upgrade to its Proteus robot marks a turning point in how humans and machines communicate at work. Instead of programming interfaces and specialized training, warehouse workers can now just talk to their autonomous colleagues—and the robots understand.
Amazon announced an upgraded version of its Proteus autonomous mobile robot with natural-language understanding capabilities, allowing warehouse workers to issue commands without programming experience. The company plans to deploy Proteus across Europe starting in the first half of 2025, alongside expanding deployments of its Vulcan picking robot and STARK collaborative tote-handling system, as part of a €10+ billion modernization investment in European fulfillment operations.
Amazon's AGI group presents a new "audit-then-score" protocol for evaluating AI-generated research reports, addressing the challenge that traditional static benchmarks fail for complex AI evaluation tasks. The approach uses AI models to actively scrutinize and refine human-generated benchmarks, improving fact-checking accuracy from 60.8% to 90.9%, and reframes ground truth as an evolving process rather than a fixed dataset.
Amazon announced 68 research award recipients from 49 universities across 11 countries for its Fall 2025 cycle, funding research in areas including AI for information security, agentic AI, automated reasoning, cryptography, cybersecurity, and sustainability. Award recipients gain access to Amazon's datasets and AWS AI/ML services, with Amazon providing consultation and collaboration opportunities to advance research in threat detection, AI safety, and abuse prevention.
The Robot Report Podcast features venture capitalist Ajay Agarwal from Bain Capital Ventures discussing his investment thesis in robotics and AI, drawing on his experience backing early-stage companies including Kiva Systems (now Amazon Robotics) and firms focused on industrial automation, supply chains, and humanoid robotics. The episode explores what makes successful robotics founders effective systems thinkers and covers recent news including Hello Robot's Stretch 4 and Automated Tire's autonomous tire-changing technology.
Amazon researchers introduce set-supervised fine-tuning (SSFT) and global forking policy optimization (GFPO) to train large language models that generate diverse reasoning paths. The approach uses global forking tokens to elicit distinct reasoning strategies, achieving 5-7% improvements in single-shot accuracy on benchmarks like AIME 2025 and LiveCodeBench while avoiding mode collapse where different reasoning modes become identical.
Colin Angle, co-founder of iRobot, has launched Familiar Machines & Magic, a startup developing consumer companion robots called Familiars designed to form emotionally intelligent, long-term relationships with users. The company, which emerged from stealth, combines Angle's 30+ years of consumer robotics experience with a team including veterans from iRobot, Amazon, Disney Research, MIT, and Boston Dynamics to build a platform for artificial life.
Amazon engineers developed advanced optimization tools combining machine learning and Monte Carlo methods to design resilient middle-mile delivery networks that can adapt to uncertainty. The approach stress-tests network designs across hundreds of plausible scenarios rather than optimizing for a single forecast, enabling the company to handle disruptions like demand spikes and logistical delays while maintaining same-day and next-day delivery promises.
Agentic mechanism enables Amazon and vendors to optimize supply chain management without disclosing private information.
Amazon, Microsoft, NVIDIA, OpenAI, Google, and xAI have all signed deals to provide AI to the Pentagon's classified networks. The speed and scale of this shift raises uncomfortable questions about who's really driving AI development—and for what purpose.
Amazon scientists and policy experts outline the company's responsible AI (RAI) pipeline, which integrates safety, fairness, and accountability throughout the AI development lifecycle across four phases: pretraining, post-training, evaluation, and frontier-risk assessment. The approach is supported by over 70 internal and external RAI tools, 500+ research papers, and tens of thousands of hours of employee training, guided by eight core pillars including safety, fairness, privacy, and transparency.
Amazon Web Services, Microsoft, NVIDIA, and Reflection AI have signed agreements to provide AI technologies to the Pentagon for use on classified military networks, joining OpenAI, Google, and xAI in similar deals. The announcements highlight the rapid adoption of AI by the US Defense Department, though Anthropic remains the only major US AI provider without a Pentagon agreement after refusing to remove safeguards on its Claude chatbot.
Amazon researchers demonstrate three attacks that can extract private training data from AI models—membership inference, gradient-based reconstruction in federated learning, and data extraction from shared models—and show how differential privacy and secure multiparty computation provide effective cryptographic defenses against these threats.
OpenAI and AWS are expanding their strategic partnership to bring OpenAI models (including GPT-5.5), Codex, and Amazon Bedrock Managed Agents to AWS customers. The integration allows enterprises to access OpenAI's frontier models and agentic capabilities within their existing AWS infrastructure, security protocols, and workflows, with all three offerings launching in limited preview.
Amazon researchers and UIUC colleagues introduced the C3LLM framework, a statistical method for assessing catastrophic failure risks in large language models during multi-turn conversations. The framework uses graph-based modeling and Clopper-Pearson confidence intervals to compute probabilistic bounds on attack success rates, addressing limitations of traditional red-teaming approaches that focus on isolated prompts rather than conversational contexts.
Google plans to invest up to $40 billion into AI startup Anthropic, with $10 billion upfront and $30 billion contingent on performance milestones, alongside a commitment to provide 5 gigawatts of computing capacity using Google's TPU infrastructure. This investment follows a similar $25 billion deal Anthropic announced with Amazon in April, exemplifying the circular investment and partnership patterns common in the competitive AI industry.
Mikell Taylor, head of robotics strategy at GM, will lead a Women in Robotics Breakfast at the 2026 Robotics Summit & Expo in Boston to discuss trust and mentorship in the field, where women represent only 19% of robotics engineers. Taylor, who previously led Amazon Robotics' development of the Proteus autonomous mobile robot, will also deliver a keynote on robot design and trustworthiness at the event.
Amazon announced a $5 billion immediate investment in Anthropic with up to $20 billion in additional milestone-based payments, following previous investments in 2023 and 2024. As part of the deal, Anthropic committed to spending over $100 billion on AWS technologies over the next decade and securing up to 5 gigawatts of chip capacity, while Anthropic's Claude AI platform will be integrated into AWS for customer access.
The RBR50 Gala will return at the 2026 Robotics Summit & Expo on May 27 in Boston, honoring top robotics innovators and showcasing award winners including Amazon's Vulcan warehouse robot, Physical Intelligence's robot learning models, and Harvard's soft exoskeleton. The event will feature over 50 technical sessions, 70+ speakers from leading robotics and AI companies, and networking opportunities across artificial intelligence, design, healthcare, and logistics applications.
Three separate announcements in the past week signal a dramatic shift: AI is no longer an experimental tool in pharmaceutical research — it's becoming the primary infrastructure. From OpenAI's specialized GPT-Rosalind to Amazon's molecular prediction models and Johns Hopkins' antibody benchmark, the industry is racing to claim territory in what may be biotech's most consequential transformation.
Amazon's Generative AI Innovation Center developed customized Amazon Nova language models fine-tuned for molecular-property prediction in drug discovery, demonstrating that a single optimized LLM can match the accuracy of multiple specialized graph neural networks while providing improved usability and reasoning capabilities for medicinal chemists. The approach combines supervised and reinforcement fine-tuning to create a unified AI assistant that simplifies workflows and accelerates drug discovery by enabling chemists to query multiple molecular properties in one interaction rather than managing separate models.
OpenAI's paused UK data center and Amazon's delayed satellite internet reveal an uncomfortable pattern: the AI industry's grand infrastructure promises keep hitting the same wall of energy costs and regulatory friction. When deployment consistently trumps development, something's broken.
Amazon's satellite-based internet service, Leo, will enter service by mid-2026, so says company CEO Andy Jassy. Writing in his annual letter, Jassy claimed Leo would offer download speeds of up to 1Gb...
Amazon CEO Andy Jassy outlined the company's robotics strategy in its 2026 shareholder letter, emphasizing robotics as key to faster delivery and lower costs, with over 1 million robots currently operating in fulfillment centers. The company recently acquired RIVR (quadruped delivery robots) and Fauna Robotics (humanoid robot developer) and plans to invest $4 billion in rural delivery expansion and pursue drone delivery through Prime Air, aiming to serve 30 million customers and deliver 500 million packages by decade's end.
Amazon Leo (formerly Project Kuiper) will launch mid-2026, according to Andy Jassy. On Wednesday, the Amazon CEO dropped the news in his annual letter to shareholders. The company says Leo will suppor...
Amazon has developed RuleForge, an agentic AI system that generates cybersecurity detection rules 336% faster than manual methods while maintaining high precision. The system uses specialized AI agents to decompose rule creation into stages (ingestion, generation, evaluation, validation) and employs a judge model that reduces false positives by 67%, enabling security teams to respond to new vulnerabilities at scale.
Amazon's acquisition of Fauna Robotics appears to be a strategic platform play focused on humanoid robot development rather than a consumer robotics venture. Fauna's Sprout humanoid robot, standing 1.07 meters tall with 29 degrees of freedom, emphasizes developer accessibility and real-world reliability through modular control systems, VR teleoperation, and constrained learned control policies rather than ambitious end-to-end autonomy, demonstrating a practical approach to humanoid development from founding in early 2024 to acquisition in March 2026.
The Trump administration's labor board has ordered Amazon to recognize and bargain with the International Brotherhood of Teamsters union, which represents workers at a warehouse in Staten Island. This...
Amazon researchers describe techniques for improving LLM-based text-to-speech systems, addressing accent leakage in multilingual synthesis, enhancing expressiveness through classifier-free guidance, and improving robustness using chain-of-thought reasoning. The work demonstrates 5-20% quality improvements across nine locales in English, French, Italian, German, and Spanish using low-rank adaptation and data augmentation methods.
This roundup highlights the top 10 robotics developments from March 2026, featuring major announcements including humanoid robot debuts at industry conferences, acquisitions by Amazon (Fauna Robotics and RIVR), new humanoid startups like Noble Machines and Rhoda AI, and industrial applications such as BMW's deployment of Hexagon's wheeled humanoids in manufacturing. The month was marked by significant activity at Smart Factory & Automation World and NVIDIA GTC, showcasing advances in humanoid robots, perception systems, robotic manipulation, and physical AI for production-scale applications.
DoorDash, Amazon, and FedEx are all betting on autonomous delivery vehicles that look nothing like traditional cars. The scramble to automate last-mile logistics is creating a new category of robots designed specifically for packages, not people.
The 2026 Robotics Summit & Expo will feature a keynote panel on building reliable robots at scale, with experts from Amazon Robotics, Locus Robotics, QNX, and Universal Robots discussing architecture patterns, safety-certified processes, and real-time systems for commercial and industrial robotics. The session will cover practical approaches to designing reliable hardware and software, with speakers sharing insights from careers in collaborative automation, warehouse robotics, and AI-enabled systems.
