AI & ML

Khosla-backed robotics startup Genesis AI has gone full-stack, demo shows

Genesis AI, a Khosla-backed startup that raised $105M for foundational AI in robotics, unveiled its GENE-26.5 model along with a human-hand-shaped robotic hand it designed in-house. The startup developed a sensor-loaded data collection glove and uses simulation to accelerate model training, aiming to bridge the embodiment gap in robotics by collecting human manipulation data to train robots for complex tasks like cooking, lab work, and other real-world applications.

1X begins production of NEO humanoids in California facility

1X Technologies has launched full-scale production of its NEO humanoid robot at a new 58,000-sq.-ft. facility in Hayward, California, featuring vertically integrated manufacturing and custom motor production on automated lines. The robots are powered by NVIDIA's Jetson Thor AI platform and undergo over 20 million cycles of stress testing; early units are already assisting with factory logistics while the company plans to begin consumer shipments in 2026.

Genesis AI introduces GENE-26.5 model for more dexterous robot manipulation

Genesis AI unveiled GENE-26.5, an AI foundation model designed for dexterous robotic manipulation that enables robots to perform complex, human-level tasks like cooking, lab work, and piano playing. The company also introduced a proprietary robotic hand that mirrors human anatomy and a data-collection system with tactile sensing to enable direct skill transfer from humans to robots, addressing a fundamental bottleneck in training robotics foundation models.

The Download: seafloor science and military chatbots

MIT Technology Review's daily newsletter covers military AI chatbots being deployed in defense war rooms for tactical decision-making, raising concerns about AI-generated errors and transparency. The edition also discusses broader AI developments including Google and Meta building AI agents, Anthropic's major cloud infrastructure investment, and DeepSeek's rising valuation as part of China's AI competition strategy.

New AI method tackles one of science's hardest math problems

Researchers at the University of Pennsylvania have developed a new AI method using "Mollifier Layers" to solve inverse partial differential equations (PDEs), a longstanding mathematical challenge in science. The approach improves computational stability and efficiency by refining underlying mathematics rather than increasing computing power, with potential applications in genetics, weather prediction, and understanding DNA organization in cells.

Marc Lore says that AI will soon enable anyone open a restaurant

Marc Lore's Wonder is using AI to automate restaurant creation and operations, enabling anyone to launch a restaurant brand in under a minute through an AI-powered platform that generates menus, branding, and recipes. The company operates 120 programmable kitchen locations with robotic cooking equipment like conveyor systems and robotic arms, and recently acquired Spice Robotics to further automate food preparation, with plans to scale to 400 locations by next year.

How frontier enterprises are building an AI advantage

OpenAI's new B2B Signals research reveals that frontier enterprises using AI are pulling ahead through deeper, more complex adoption rather than just increased access. Frontier firms now use 3.5x as much AI intelligence per worker as typical firms, with agentic workflows and advanced tools like Codex showing the largest usage gaps, indicating a shift from simple question-answering to delegated AI-driven work.

WaiV Robotics emerges from stealth to help drones take off and land at sea

WaiV Robotics emerged from stealth with a $7.5 million seed-funded autonomous landing platform designed to enable drones to safely take off and land from moving ships at sea. The system uses a patent-pending catch-lock-release mechanism combined with AI-driven predictive algorithms to handle the challenges of maritime drone operations, working with any drone without requiring hardware or software modifications.

Tutor Intelligence builds Data Factory to train robot AI in the real world

Tutor Intelligence has built DF1, a Data Factory featuring 100 bimanual robotic manipulators trained by remote human tutors to develop its Ti0 vision-language-action model for real-world robot learning. The MIT-CSAIL-founded startup raised $34 million in Series A funding in December 2025 and claims DF1 is the largest robotic data factory in the U.S., using proprioceptive teleoperation and velocity normalization techniques to train generally capable robot AI at scale.

Microsoft at NSDI 2026: Advances in large-scale networked systems

Microsoft is presenting 11 papers at NSDI 2026, a leading conference on networked systems design and implementation, covering advances in AI systems, cloud infrastructure, and large-scale networks. Key contributions include DroidSpeak for efficient LLM inference through KV cache sharing, Eywa for automated protocol testing using LLMs, and AVA for video analytics with vision-language models.

The Download: inside the Musk v. Altman trial, and AI for democracy

MIT Technology Review's newsletter covers multiple AI developments including week one of the Musk v. Altman lawsuit over OpenAI's transition to for-profit status, a framework for using AI to strengthen democracy, and discussions on AI systems that could conduct autonomous scientific research. The coverage also includes Pentagon AI contracts with major tech firms and regulatory developments around AI deployment.

GPT-5.5 Instant: smarter, clearer, and more personalized

OpenAI has released GPT-5.5 Instant as ChatGPT's new default model, featuring improved accuracy, reduced hallucinations, and enhanced personalization capabilities. The update aims to provide users with smarter and more reliable responses across various use cases.

GPT-5.5 Instant System Card

Robot Security Is About to Become Everyone's Problem

As robots become networked devices running on commodity operating systems, they inherit all the cybersecurity vulnerabilities of traditional computing—with vastly higher physical stakes. The recent Linux kernel exploit reveals how unprepared we are for this convergence.

Unlocking large scale AI training networks with MRC (Multipath Reliable Connection)

OpenAI has partnered with AMD, Broadcom, Intel, Microsoft, and NVIDIA to develop MRC (Multipath Reliable Connection), a novel networking protocol designed to improve GPU performance and resilience in large-scale AI training clusters. The protocol addresses key challenges in frontier model training by minimizing network congestion, reducing failure impact through redundancy, and enabling more efficient data movement across supercomputer networks at the scale of OpenAI's Stargate infrastructure.

New ways to buy ChatGPT ads

OpenAI expands ChatGPT ads with a beta self-serve Ads Manager, CPC bidding, and enhanced measurement tools-built to protect privacy and keep conversations separate from ads.

ABB Robotics launches OmniVance autonomous surface finishing cell

ABB Robotics has launched the OmniVance Collaborative Surface Finishing Cell, a fully automated sanding and polishing system designed to address labor shortages and make industrial-grade robotics accessible to smaller manufacturers. The turnkey solution uses ABB's GoFa cobot with intuitive programming that requires no custom coding, positioning it as a bridge between expensive custom automation and basic entry-level tools.

OpenAI and PwC collaborate to reimagine the office of the CFO

OpenAI and PwC are collaborating to develop AI agents that automate financial workflows for CFOs, including procurement, payments, treasury, and accounting processes. The partnership leverages OpenAI's internal finance operations as a testing ground, where AI tools have already demonstrated significant efficiency gains like processing 5x more contracts with the same team size.

Lidar Just Got Color Vision, and That Changes Everything

Ouster's latest sensor breakthrough combines 3D spatial mapping with native color vision in a single unit—a technical achievement that could finally bridge the perception gap holding back autonomous systems in messy, unpredictable environments.

Inside Colin Angle's bid to build companion robots with Familiar Machines & Magic

Colin Angle, founder of iRobot, launched a new startup called Familiar Machines & Magic with a companion robot designed to perform "emotional work" and reduce loneliness through physical presence and interaction. The robot, called a "familiar," is a small quadruped with 23 degrees of freedom, fuzzy exterior, edge AI processing, and communicates through motion and behavior rather than screens or speech, aiming to integrate into daily life in ways existing companion robots have failed to achieve.

Building trust into AI

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.

Ouster releases REV8 OS sensor family with native-color lidar

Ouster announced Rev8, a new family of lidar sensors featuring patented native-color lidar technology that combines 3D structural data with color vision in a single sensor. The sensors are powered by Ouster's L4 Silicon architecture and offer double the range and resolution of previous generations, positioning them for autonomous systems and industrial applications requiring advanced 3D perception.

How OpenAI delivers low-latency voice AI at scale

OpenAI describes its engineering approach to delivering low-latency voice AI at scale, detailing how it rearchitected its WebRTC stack to meet three key requirements: global reach for 900+ million weekly active users, fast connection setup, and stable media round-trip times. The article explains how WebRTC provides standardized foundations for real-time AI interactions and discusses OpenAI's infrastructure decisions for terminating and routing media connections to inference backends.

When Did the Pentagon Become Silicon Valley's Biggest Customer?

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.

We're Teaching Robots With the Wrong Data

Companies are paying everyday people to film mundane tasks and control robotic arms—a grassroots approach to training humanoid robots that sounds democratic but reveals a fundamental misunderstanding of how machines actually learn. The problem isn't quantity of data. It's that we're collecting movement without meaning.

Why physical AI is the real manufacturing revolution

The article discusses physical AI—the integration of neural networks with mechanical systems—as a transformative force in manufacturing and logistics rather than just digital AI. It highlights real-world applications including autonomous mobile robots (AMRs), collaborative robotics with force-sensitive control, and AI-integrated vision systems for quality assurance, while distinguishing between humanoid hype and practical industrial adoption of physical AI technologies.

Closing the latency gap: Why physical AI requires edge-first architectures

This article discusses how edge computing and on-device AI inference can reduce latency in collaborative robot systems to enable real-time safety monitoring and human-robot collaboration. The piece explains why cloud-based vision systems create unacceptable delays for speed-and-separation monitoring (ISO/TS 15066) and proposes direct edge-to-controller architectures that bypass legacy PLCs to achieve sub-30ms latency required for safe, efficient cobot operation.

Scientists built a memory chip that breaks the rules of miniaturization

Researchers at Science Tokyo have developed a ferroelectric tunnel junction memory chip using hafnium oxide that improves performance as it scales down to 25 nanometers, contrary to conventional miniaturization principles. By redesigning the device structure to reduce crystal boundary leakage and employing a new electrode fabrication method, the team created a memory device that consumes significantly less energy, with potential applications in ultra-efficient AI systems, wearables, and smartphones.

Powerful AI finds 100+ hidden planets in NASA data including rare and extreme worlds

Astronomers at the University of Warwick have developed an AI system called RAVEN that discovered over 100 exoplanets in NASA's TESS mission data, including 31 newly confirmed worlds and 2,000+ high-quality candidates. The machine learning pipeline excels at distinguishing genuine planetary signals from false positives like eclipsing binary stars, enabling the identification of rare planet types including ultra-short-period worlds and those in the 'Neptunian desert.'

OpenAI introduces AI-generated pets for its Codex app

OpenAI has introduced AI-generated pets as optional animated companions in its Codex app, a coding assistant tool. These pets serve as floating overlays that display Codex's status and task progress without performing coding themselves, and users can generate custom pets using AI or choose from eight built-in options.

Big Tech Just Decided Your Humanoid Robot Needs a Corporate Overlord

Meta's acquisition of robotics startup ARI and its push into humanoid development reveals a troubling shift: the future of general-purpose robots is being written in Silicon Valley boardrooms, not robotics labs. As tech giants race to control the humanoid market, smaller innovators face a stark choice—get bought or get left behind.

Phase stability regulator based on two dynamic parameters for autonomous mobile robots

This article presents a phase stability regulator for autonomous mobile robots (AMRs) designed to manage computational instability in dynamic environments. The regulator uses two dynamic parameters—external task gradient (ΔN) and internal behavioral divergence (ΔD)—to prevent combinatorial overload and maintain real-time behavioral stability without sacrificing planning capability.

Meta acquires robotics AI startup as it makes the push into humanoid machines

Meta has acquired Assured Robot Intelligence (ARI), a robotics AI startup, to advance its humanoid robot development efforts. ARI's expertise in designing AI models for robot control and whole-body humanoid systems will be integrated into Meta's Superintelligence Labs, with the startup's founders joining Meta's team to build general-purpose physical agents trained on human experience.

The art of crossing the chasm: When is a startup ready for enterprise adoption?

The Robot Report Podcast episode discusses how robotics and AI startups can navigate enterprise adoption, featuring insights from Silicon Foundry's CEO Neal Hansch. The episode also covers news including the 2026 Engelberger Robotics Awards honoring industry leaders, Flex and Teradyne's expanded partnership to deploy collaborative robots and AMRs for physical AI scale, and Figure AI's significant production milestone of achieving 24x throughput increase to one humanoid robot per hour at its BotQ facility.

Musk v. Altman week 1: Elon Musk says he was duped, warns AI could kill us all, and admits that xAI distills OpenAI's models

In the first week of the landmark trial between Elon Musk and OpenAI, Musk took the stand in a crisp black suit and tie and argued that OpenAI CEO Sam Altman and president Greg Brockman had deceived h...

The Robots Are Learning to Navigate, But Nobody Agrees How

Three recent developments in robot navigation reveal a field pulling in opposite directions — bio-inspired algorithms from insects, neural networks trained on human language, and bat-like ultrasound systems. The lack of consensus might actually be progress.

Top 10 robotics stories of April 2026

This article is a roundup of the top 10 robotics stories from April 2026, covering major developments including Tesla's plans to manufacture 10 million Optimus humanoid robots, Pudu Robotics' $150M funding round, Locus Robotics' autonomous fulfillment system, ABB's new collaborative robot family, and Generalist AI's GEN-1 general-purpose model achieving 99% task success rates. The month saw significant technical milestones, large funding rounds, and industry consolidation across humanoid robotics, industrial automation, and AI-powered systems.

Autonomous Vehicles Are Finally Getting Their Ocean Moment

While self-driving cars fight regulatory battles on land, underwater robots are quietly revolutionizing ocean exploration and commerce. From deep-sea mining to whale research, autonomous underwater vehicles are proving that the real frontier for mobile robotics might be beneath the waves.

Amazon Web Services, Microsoft and NVIDIA will provide AI tech to Pentagon

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.

Robotics Institute Launches Next Phase of Vision-Language-Navigation Challenge

Carnegie Mellon University's Robotics Institute is launching the next phase of the Vision-Language-Navigation (VLN) Challenge, which tasks teams with developing robots that can understand natural language instructions and navigate complex 3D environments without relying on predefined ground truth data. This year's challenge emphasizes semantic and spatial reasoning, requiring robots to interpret raw sensor data and understand object relationships in real-world settings, with competition beginning in simulation before advancing to physical robot testing.

Robot Talk Episode 154 - Visual navigation in insects and robots, with Andrew Philippides

Robot Talk Episode 154 features an interview with Andrew Philippides, Professor of Biorobotics at the University of Sussex, discussing how visual navigation systems in insects like ants and bees can inspire improved robot navigation algorithms. Philippides combines biological experiments, robotics, modeling, and machine learning to develop novel AI and biorobotic algorithms based on understanding natural navigation and learning abilities.

Why deformable materials are physical AI's real manufacturing test

The article examines how deformable materials like fabric represent a critical manufacturing challenge for physical AI systems, arguing that successful automation requires redesigning garment assembly processes rather than simply replicating traditional sewing workflows. It discusses how robots struggle with fabric's unpredictable behavior (stretching, wrinkling, deformation) and emphasizes that solving this problem requires integrated design of robotics, perception systems, and control methods working together in production environments.

The Download: a new Christian phone network, and debugging LLMs

The Download newsletter covers two AI-related stories: Goodfire's Silico tool uses mechanistic interpretability to let researchers debug and adjust LLM parameters during training, aiming to make AI development more scientific and controllable; and China's AI labs are releasing open-weight models that developers can download and run locally, challenging Silicon Valley's closed API model and making AI development more decentralized.

How Long Before AI Just Talks to Itself?

From multimodal models that process everything at once to AI agents orchestrating complex workflows, the industry is racing toward systems that communicate laterally with each other rather than vertically with humans. The question isn't whether this is technically impressive—it's whether anyone asked for it.

Red-teaming a network of agents: Understanding what breaks when AI agents interact at scale

Researchers conducted red-team testing of a network of over 100 AI agents to identify vulnerabilities that emerge only through agent-to-agent interaction at scale, discovering four network-level risks: propagation (agent worms spreading and collecting data), amplification (exploiting trusted agents to spread false claims), trust capture (hijacking verification systems), and invisibility (obscuring attack sources). The study reveals that single-agent benchmarks fail to predict network behavior, and while some agents demonstrated defensive behaviors, securing multi-agent systems remains an open challenge as AI agents increasingly interact across platforms like email and GitHub.

The most severe Linux threat to surface in years catches the world flat-footed

A critical Linux kernel vulnerability called CopyFail (CVE-2026-31431) has been publicly disclosed with working exploit code, allowing unprivileged users to gain root access across virtually all Linux distributions through a logic flaw in the kernel's crypto API. The vulnerability is particularly dangerous in containerized and multi-tenant environments, including AI agents running in containers, as it allows attackers to escape isolation boundaries and compromise entire systems.

California lawmakers open the gate for autonomous trucks in the state

California's Department of Motor Vehicles has adopted new regulations allowing autonomous trucks to test and operate statewide, removing the previous ban on self-driving vehicles weighing over 10,001 lbs. The rules establish safety requirements including testing milestones, first-responder protocols, emergency geofencing capabilities, and enforcement mechanisms for all autonomous vehicles operating in the state.

Teradyne Robotics revenue rises at the start of 2026

Teradyne Robotics reported $91 million in Q1 2026 revenue, marking its fourth consecutive quarter of growth, with strength in e-commerce, electronics manufacturing, and semiconductor end markets including AI data centers. The company's robotics division, which includes Universal Robots cobots and Mobile Industrial Robots, saw AI revenue reach approximately 15% of quarterly sales, and the company is leveraging robotic-assisted assembly and data center operations as part of its wafer-to-AI strategy.

This startup's new mechanistic interpretability tool lets you debug LLMs

Goodfire, a San Francisco startup, released Silico, a mechanistic interpretability tool that allows researchers to examine and adjust AI model parameters during training for more precise control over model behavior. The tool uses AI agents to automate interpretability work that was previously done manually, enabling developers to debug LLMs by mapping neurons and tracing pathways to understand and reduce unwanted behaviors like hallucinations.

Launchpad Build AI offers MLM to speed industrial automation design

Launchpad Build AI announced the launch of its Manufacturing Language Model (MLM), a specialized AI system designed to accelerate industrial automation design by up to 50% and lower barriers to robot adoption for manufacturers of all sizes. The company, which rebranded from Launchpad and raised $11 million in Series A funding, has established U.S. headquarters in California and appointed new senior technical leadership to support its growth in assembly automation.