autonomousdevice.ai

#Autonomous Device AI Meta

#AI models deployed in embedded systems at edge | Brushless DC motors | Hall effect sensors | Optical encoders | Sensorless motor control | Field-oriented control | Artificial intelligence at edge | Three fundamental modalities: vision, sound, and motion | Using AI models to infer information about device environment | Linear algorithms | Software and hardware combination | Deploying multiple AI models in embedded devices requires edge processors designed to run AI | Embedded systems using AI can be considered open | Sensor fusion utilizes combined data from multiple sensors | AI-based vision systems are more adaptable to natural variations inherent in object inspection | Objects can be identified and inspected more quickly with greater flexibility | Strong multimodal AI, a single model will process multiple types of data | Control algorithms will use inputs generated by AI, inferred from multiple sources of data | AI inferencing in data flow | AI-enabled image sensors are perfect for gesture detection | Event detection based on sound is an active area of development | On device learning in real time

#Edge AI | Ultra-low power wireless connectivity for IOT-sensors | Eye-tracking | Gesture recognition | Object detection with cameras and sensors | Real time lidar intelligence | ECG Anomaly Detection | Neuromorphic Edge AI accelerator co-processor chip delivering exceptional performance with minimal power consumption

#SB53 | Law requires large AI model developers to publish frameworks on their websites including how company responds to critical safety incidents, assesses, manages catastrophic risk | Companies must report critical safety incidents to CA state within 15 days, or within 24 hours if a risk believed to pose an imminent threat of death or injury | Addressing catastrophic risk posed by advanced AI models, called frontier models | Law addresses risks in the context of an operator losing control of an AI system | Transparency report must include intended uses of a model, restrictions or conditions of using a model, how a company assesses and addresses catastrophic risk, and whether those efforts were reviewed by an independent third party | Rishi Bommasani, Stanford University, consulted Gov. Gavin Newsom | Excluded impact of AI systems on environment | Only targets companies that make $500 million in annual revenue | Excluded information companies characterize as trade secrets (common way to prevent sharing information about AI models) | Office of Emergency Services will produce anonymized report about critical safety incidents

#Ability to change resource usage over time

#Off road Autonomy

#Industrial AI | Robotics | Simulation | Edge Computing Ecosystems | Humanoid robots | Scaling robot automation beyond isolated workcells | AI factories | Digital twins | AI-driven design | Mobile robots | AI in semiconductor industry

#Autonomous Inspection

#Robotic AI

#Robotic Autonomy

#Uncertainty aware AI

#Handling tasks without human intervention

#Spontaneously learning and improving from experiences

#Delivery robot

#Equip On Your Existing Device

#Precision application

#Manual use switch

#Targeted Efficiency

#Optimizing resources

#Token | Numerical representations of words and characters | LLMs take tokens as input | LLMs generate tokens as output | Input text is translated into tokens by a tokenizer | Different LLMs use different tokenizers

#Tokenizer | Translates Input text into tokens | Different LLMs use different tokenizers | Token is numerical representations of words and characters | LLMs take tokens as input | LLMs generate tokens as output

#Smart Insights

#Smart Decisions

#Autonomous technology

#Autonomous Platform

#Reducing labor costs

#Reducing maintenance costs

#Autonomous system

#Remotely monitoring

#Autonomous tractor

#Equipment availability

#Equipment runtime

#Dynamic sensing

#Bringing sensors to assets

#Dispatching agile mobile robots equipped with sensors to collect data on site

#More accurate data models

#Virtual Wall Functionality

#Automatic Movement Control

#Camera operating as an Edge Device

#Compressed video stream (H264)

#RTSP protocol

#MJPEG stream

#Live image with AI overlay

#Intelligent camera

#Digital Manufacturing

#Autonomous routes

#Digital Twin

#Points of Interest (POI)

#Augmented reality

#3D camera

#Z-accuracy

#Stereo camera

#Field of view

#RGB sensor

#AI-in-a-box

#Neuromorphic processor

#Sustainable AI technology

#Low-shot learning

#Biometric recognition

#Image classification model

#Temporal Event Neural Network

#AI and machine learning (ML) for EV design and validation | Short-term memory (LSTM) network | Dynamic Mode Decomposition (DMD) | ROM | Recurrent Neural Network (RNN)

#Intelligent sensor technology

#Natural hydrogen: < $1/kg

#White hydrogen: carbon intensity with 75% hydrogen and 22% methane, rises to 1.5 kg CO2e per kg H2

#Gold hydrogen: highest production tax credits (PTC) because the lifecycle carbon intensity below 4 kg CO2e per kg H2

#Grey hydrogen: produced from fossil fuels, costs less than $2 per kilogram (kg) of hydrogen on average

#Green hydrogen: > $6/kg

#Machine learning (ML) to reliably distinguish between a significant and insignificant event on the power grid

#Generative physical AI

#Robotics stack

#Humanoid foundation model

#Autonomous machine development

#Cognitive AI-driven capabilities

#Robot-agnostic software platform

#Robotic-grasping

#Synthetic data generation

#Autonomous mobile robot

#Digital twin technology

#Reference workflow

#Training robots in virtual environment

#Robotic arm

#Robot work cell

#IoT data

#Digital twin workflow

#AI-enabled autonomous machine

#Software frameworks and robot model

#Human coworker

#Intelligent assistant

#Edge AI solution

#Robotics skills

#Electric Vehicle (EV) charging

#Open Smart Charging Protocol (OSCP)

#Cell-to-pack (CTP) design: integrating battery cells directly into battery pack

#Cell-to-chassis (CTC) design: incorporating battery cells directly into vehicle chassis

#Skateboard platforms pre-equipped for self-driving capabilities

#10-minute EV charging benchmark

#Perceptual AI hardware

#Perceptual AI software

#Perceptual AI algorithm

#Neural network architecture

#AI vision system

#Device-agnostic AI system

#Transformers: class of neural network models originally designed for natural language processing

#Polynomial-based continuous convolutions

#Neuromorphic sensing and computing

#Spiking neural network algorithms

#AI Processing Unit (AIPU)

#Vision AI in (biometric) access control

#8-bit integer inference arithmetic vs 32-bit floating-point full-precision

#False positives

#False negatives

#Temporal Event Neural Networks (TENNs)

#Capturing thermal, acoustic, and visual data consistently

#Cardiac ablation: medical procedure used to treat irregular heart rhythms (arrhythmias) by creating small scars in heart tissue

#Atrial flutter: abnormal heart rhythm

#Fluoroscopy: medical imaging technique using X-rays to create real-time moving images of internal structures such as heart

#Active MR tracking: real-time localization of catheters during MRI-guided interventions, utilizing microcoils or antennas to provide precise positional information, enhancing visualization within MR images, reducing the need for manual adjustments and improving efficiency compared to passive methods, integrates tracking sequences with imaging, allowing for dynamic updates of the imaging plane as the device moves

#Microcoils: enhancing MR tracking accuracy by providing improved sensitivity and localization of devices within MRI environment, enabling high frame rates, allowing real time tracking of moving instruments

#Micro transmit tracking

#Centroid pixel method

#Phase field Dithering

#Active microcoils

#Automatic registration of tracked devices

#IO-Link: an open-standard communication protocol (IEC 61131-9) designed for connecting sensors and actuators in industrial automation

#Detecting anomalies

#Active stereo vision

#Time Of Flght (TOF)

#3D depth sensing

#Smart home device

#Miltimodal

#Perceptual mode

#1550nm LiDAR | Advantages: safety, range, and performance in various environmental conditions | Enhanced Eye Safety: absorbed more efficiently by cornea and lens of eye, preventing light from reaching sensitive retina | Longer Detection Range | Improved Performance in Adverse Weather Conditions such as as fog, rain, or dust | Reduced Interference from Sunlight and Other Light Sources | More expensive due to complexity and lower production volumes of their components

#SLAM | Simultaneous Localization and Mapping | Allows robots and autonomous systems to build a map of an unknown environment while estimating their own motion within it | Visual Al technology | Real-time tracking of both manual and autonomous fleets | Every movement transformed into actionable data | Collision prevention and predictive risk alerts | Optimising operations and empowering workers and robots to move in sync with real-time shared spatial awareness | Main components: front-end matching, state estimation, map management and loop closure | Design trade-offs: compute versus accuracy | Camera-centric systems challenges: fog, rain, snow, glare, low light | Radar based SLAM as complementary or alternative for building maps and estimating position under adverse conditions

#Building Information Modeling (BIM)

#Architecture, Engineering, and Construction (AEC)

#3D modeling

#4D modeling (time scheduling)

#5D modeling (cost estimation)

#SLice Integration by Vision Transformer (SLIViT)

#Retinal scan

#Ultrasound video

#CT

#MRI

#Disease-risk biomarker

#Pre-training method

#Fine-tuning method

#Disease trajectory

#Tailoring treatment

#Fine-tuning 2D model on 3D scans

#Downstream learning

#Deep-learning framework

#Agentic workflow

#Vector database

#Learning Management System (LMS)

#Time To First Token (TTFT)

#Multimodal AI

#Robotic embodiment

#Humanoid robot

#Universal Scene Description (OpenUSD)

#Cognitive robotics

#Vertical movement

#Actuated linear guide system

#Actuator

#Abstraction and Reasoning Corpus | Dataset created by François Chollet to evaluate gap between human and machine learning capabilities | 1,000 image-based reasoning tasks requiring abstract thinking and reasoning skills | Tasks involve identifying patterns and applying them to generate correct outputs

#Corpus | Structured collection of texts or language data used for linguistic research

#Resistive RAM (ReRAM) technology | onsemi Treo platform to provide embedded non-volatile memory | ReRAM integration into Bipolar CMOS DMOS (BCD) process | Potential alternative to flash memory | Demand for faster, more efficient, and scalable memory solutions increasing | Lower power consumption | Less vulnerable to common hacking tactics | ReRAM can be integrated easily into chip designs without interfering with power analog components

#California wildfire | Challenges | Access roads too steep for fire department equipment | Brush fires | Dangerously strong winds for fire fighting planes | Drone interfering with wildfire response hit plane | Dry conditions fueled fires | Dry vegetation primed to burn | Faults on the power grid | Fires fueled by hurricane-force winds | Fire hydrants gone dry | Fast moving flames | Hilly areas | Increasing fire size, frequency, and susceptibility to beetle outbreaks and drought driven mortality | Keeping native biodiversity | Looting | Low water pressure | Managing forests, woodlands, shrublands, and grasslands for broad ecological and societal benefits | Power shutoffs | Ramping up security in areas that have been evacuated | Recoving the remains of people killed | Retardant drop pointless due to heavy winds | Smoke filled canyons | Santa Ana winds | Time it takes for water-dropping helicopter to arrive | Tree limbs hitting electrical wires | Use of air tankers is costly and increasingly ineffective | Utilities sensor network outdated | Water supply systems not built for wildfires on large scale | Wire fault causes a spark | Wires hitting one another | Assets | California National Guard | Curfews | Evacuation bags | Firefighters | Firefighting helicopter | Fire maps | Evacuation zones | Feeding centers | Heavy-lift helicopter | LiDAR technology to create detailed 3D maps of high-risk areas | LAFD (Los Angeles Fire Department) | Los Angeles County Sheriff Department | Los Angeles County Medical Examiner | National Oceanic and Atmospheric Administration | Recycled water irrigation reservoirs | Satellites for wildfire detection | Sensor network of LAFD | Smoke forecast | Statistics | Beachfront properties destroyed | Death tol | Damage | Economic losses | Expansion of non-native, invasive species | Loss of native vegetation | Structures (home, multifamily residence, outbuilding, vehicle) damaged | California wildfire actions | Animals relocated | Financial recovery programs | Efforts toward wildfire resilience | Evacuation orders | Evacuation warnings | Helicopters dropped water on evacuation routes to help residents escape | Reevaluating wildfire risk management | Schools closed | Schools to be inspected and cleaned outside and in, and their filters must be changed

#IoT sensor

#A-list celebrity home protector | Burglaries targeting high-end items | Burglary report on Lime Orchard Road | Burglar had smashed glass door of residence | Ransacked home and fled | Couple were not home at the time | Unknown whether any items were taken | Lime Orchard Road is within Hidden Valley gated community of Los Angeles in Beverly Hills | Penelope Cruz, Cameron Diaz, Jennifer Lawrence, Adele and Katy Perry have purchased homes there, in addition to Kidman and Urban | Kidman and Urban bought their home for $4.7 million in 2008 | 4,100-square-foot, five-bedroom home built in 1965 and sits on 1¼-acre lot | Property large windows have views of the canyons | Theirs is one of several celebrity properties burglarized in Los Angeles and across country recently | Connected to South American organized-theft rings

#Professional athlete home protector | South American crime rings | Targeting wealthy Southern California neighborhoods for sophisticated home burglaries | Behind burglaries at homes of professional athletes and celebrities | Theft groups conduct extensive research before plotting burglaries | Monitoring target whereabouts and weekly routines via social media | Tracking travel and schedules | Conducting physical surveillance at homes | Attacks staged while targets and their families are away | Robbers aware of where valuables are stored in homes prior to staging break-ins | Burglaries conducted in short amount of time | Bypass alarm systems | Use Wi-Fi jammers to block Wi-Fi connections | Disable devices | Cover security cameras | Obfuscate identities

#Self-learning infrared digital anti-masking system | System continuously monitors the sensor environment and adjusts accordingly, reducing the chances of false alarms while maintaining security performance | Unlike manually adjustable ones it adapts upon installation | Anti-masking | Designed to detect and respond when sensor is deliberately covered or obstructed | Anti-masking is critical feature in professional security solutions | Achieved by emitting infrared or microwave signals that continuously check for obstructions | Allowing user to fine-tune detection sensitivity to suit environment minitored

#Artificial intelligence | Accelerated growth in space and satellite industry | Enhancing data processing, automation and decision-making | Enabling faster image analysis | Real-time satellite adjustments | Predictive maintenance | Autonomous spacecraft operations | Making space technology more efficient, cost-effective and scalable

#Dexterous robot | Manipulate objects with precision, adaptability, and efficiency | Dexterity involves fine motor control, coordination, ability to handle a wide range of tasks, often in unstructured environments | Key aspects of robot dexterity include grip, manipulation, tactile sensitivity, agility, and coordination | Robot dexterity is crucial in: manufacturing, healthcare, logistics | Dexterity enables automation in tasks that traditionally require human-like precision

#Agentic AI | Artificial intelligence systems with a degree of autonomy, enabling them to make decisions, take actions, and learn from experiences to achieve specific goals, often with minimal human intervention | Agentic AI systems are designed to operate independently, unlike traditional AI models that rely on predefined instructions or prompts | Reinforcement learning (RL) | Deep neural network (DNN) | Multi-agent system (MAS) | Goal-setting algorithm | Adaptive learning algorithm | Agentic agents focus on autonomy and real-time decision-making in complex scenarios | Ability to determine intent and outcome of processes | Planning and adapting to changes | Ability to self-refine and update instructions without outside intervention | Full autonomy requires creativity and ability to anticipate changing needs before they occur proactively | Agentic AI benefits Industry 4.0 facilities monitoring machinery in real time, predicting failures, scheduling maintenance, reducing downtime, and optimizing asset availability, enabling continuous process optimization, minimizing waste, and enhancing operational efficiency

#Field Foundation Model (FFMs) | Physical world model using sensor data as an input | Field AI robots can understand how to move in world, rather than just where to move | Very heavy probabilistic modeling | World modeling becomes by-product of Field AI.robots operating in the world rather than prerequisite for that operation | Aim is to just deploy robot, with no training time needed | Autonomous robotic systems applucations | Field AI is software company making sensor payloads that integrate with their autonomy software | Autonomous humanoid Field AI can do | Focus on platforms that are more affordable | Integrating mobility with high-level planning, decision making, and mission execution | Potential to take advantage of relatively inexpensive robots is what is going to make the biggest difference toward Field AI commercial success

#Embedded vision applications

#Edge SoC

#Heterogeneous architecture | Bbringing together CPUs, GPUs and specialized NPUs

#Thermal challenges associated with more advanced process nodes

#Programmable hardware | Delivering high-performance, high-efficiency vision applications at the edge

#Extract valuable information from camera streams

#Off-the-shelf AI model

#AI pipeline | Post-processing | Creating complete video analytics application

#AI-powered embedded vision solution

#Multipurpose commercial humanoid | Potential for useful and reliable and affordable humanoids | Difficult problem making highly technical piece of hardware and software compete effectively with humans in labor market | Robots are not hard to build; but they are hard to make useful and make money with | Whole perception pipeline running at the framerate of sensors nowadays | All the technology is here now | Starting with surrogate robot from someone else to get autonomy team going while building own robot in parallel | Giving out a significant chunk of the company to early joiners | Combined efforts of the research community enables commercialization | Building team is really important

#Infrared (IR) camera | Ability to detect and measure energy below visible light region of electromagnetic spectrum | .Useful as thermal imagers, sensing objects (and their movements) whose temperatures differ from that of ambient environment | Infrared light can also penetrate materials which visible light cannot, such as dust, fog, smoke, and thin walls | .Combining visible light and infrared data results in a richer multispectrum understanding of a scene

#Large Language Model (LLM) | Foundational LLM: ex Wikipedia in all its languages fed to LLM one word at a time | LLM is trained to predict the next word most likely to appear in that context | LLM intellugence is based on its ability to predict what comes next in a sentence | LLMs are amazing artifacts, containing a model of all of language, on a scale no human could conceive or visualize | LLMs do not apply any value to information, or truthfulness of sentences and paragraphs they have learned to produce | LLMs are powerful pattern-matching machines but lack human-like understanding, common sense, or ethical reasoning | LLMs produce merely a statistically probable sequence of words based on their training | LLMs are very good at summarizing | Inappropriate use of LLMs as search engines has produced lots of unhappy results | LLM output follows path of most likely words and assembles them into sentences | Pathological liars as a source for information | Incredibly good at turning pre-existing information into words | Give them facts and let them explain or impart them

#Retrieval Augmented Generation. (RAG LLM) | Designed for answering queries in a specific subject, for example, how to operate a particular appliance, tool, or type of machinery | LLM takes as much textual information about subject, user manuals and then pre-process it into small chunks containing few specific facts | When user asks question, software system identifies chunk of text which is most likely to contain answer | Question and answer are then fed to LLM, which generates human-language answer in response to query | Enforcing factualness on LLMs

#Large Behavior Model (LBM) | Controlling the entire robot actions | Joint research partnership between Boston Dynamics and Toyota Research Institute | Collaboration aims to create a general-purpose humanoid assistant | Whole-body movements: walking, crouching, and lifting to complete tasks that involve sorting and packing

#AI generalist robot | Developing end-to-end language-conditioned policies | Taking full advantage of capabilities of humanoid form factor, including taking steps, precisely positioning its feet, crouching, shifting its center of mass, and avoiding self-collisions | Building policies process: 1. Collect embodied behavior data using teleoperation on both real-robot hardware and in simulation, 2. Process, annotate, and curate data to easily incorporate it into machine learning pipeline, 3. Train neural-network policy using all of the data across all tasks | 4. Evaluate the policy using a test suite of tasks | Policy maps inputs consist of images, proprioception, language prompts to actions that control robot at 30Hz | Leveraging diffusion transformer together with flow matching loss to train model | Dexterous manipulation including part picking, regrasping | Subtasks triggered by passing a high-level language prompt to the policy | Reacting intelligently when things go wrong | With Large Behavior Model (LBM), training process is the same whether it is stacking rigid blocks or folding a t-shirt: if you can demonstrate it, robot can learn it | Speeding up the execution at inference time without requiring any training time changes

#Teleoperation | High-Quality Data Collection for Model Training | Control system allows to perform precise manipulation while maintaining balance and avoiding self-collisions | VR headset for operators to fully immerse themselves in the robot workspace and have access to the same information as the policy, with spatial awareness bolstered by a stereoscopic view rendered using head mounted cameras reprojected to the user viewpoint | Custom VR software provides teleoperator with a rich interface to command robot, providing them real-time feeds of robot state, control targets, sensor readings, tactile feedback, and system state via augmented reality, controller haptics, and heads-up display elements | One-to-one mapping between user and robot (i.e. moving your hand 1cm would cause robot to also move by 1cm) | To support mobile manipulation, tracking on feet added and teleoperation control extended to support stance mode, support polygon, and stepping intent to match that of operator

#Policy | Toyota Research Institute.Large Behavior Model | Diffusion Policy-like architecture | Boston Dynamic policy | Diffusion Transformer-based architecture | Flow-matching objective | Conditioned on proprioception, images | Accepting language prompt that specifies objective to robot | Image data comes in at 30 Hz | Network uses a history of observations to predict an action-chunk | Observation space consists of images from robot head-mounted cameras along with proprioception | Action space includes joint positions for left and right grippers, neck yaw, torso pose, left and right hand pose, and left and right foot poses | Shared hardware and software across two robots aids in training multi-embodiment policies that can function across both platforms, allowing to pool data from both embodiments | Quality assurance tooling allows to review, filter, and provide feedback on data collected

#Simulation | Allows to quickly iterate on teleoperation system and write unit and integration tests | Performing informative training and evaluations that would otherwise be slower, more expensive and difficult to perform repeatably on hardware | Simulation stack is faithful representation of hardware and on-robot software stack | Ability to share data pipeline, visualization tools, training code, VR software and interfaces across both simulation and hardware platforms | Benchmarking policy and architecture choices | Incorporating simulation as a significant co-training data source for multi-task and multi-embodiment policies deployed on hardware

#Vision-language model (VLM) | Training vision models when labeled data unavailable | Techniques enabling robots to determine appropriate actions in novel situations | LLMs used as visual reasoning coordinators | Using multiple task-specific models

#Robot autonomy system combining the benefits of Visual SLAM positioning with advanced AI local perception and navigation tech | Visual Al technology | AI-based autonomy solutions | Visual SLAM | Dynamic obstacle avoidance | Constructing accurate 3D maps of the environment using sensors built into robots | Algorithms precisely localize robot by matching what it observes at any given time with 3D map | Using AI driven perception system robot learns what is around it and predicts people actions to react accordingly | Intelligent path planning makes robot move around static and dynamic obstacles to avoid unnecessary stops | Collaborating with each others robots share important information like their position and changes in mapped environment | Running indoors, outdoors, over ramps and on multiple levels without auxiliary systems | Repeatability of 4mm guarantees precise docking | Updates the map and shares it with the entire fleet | Edge AI: All intelligence is on the vehicle, eliminating any issue related to the loss of connectivity | VDA 5050 standardized interface for AGV communication | Alphasense Autonomy Evaluation Kit | Autonomous mobile robot (AMR) | Hybrid fleets: manual and autonomous systems work collaboratively | Equipping both autonomous and manually operated vehicles with advanced Visual SLAM and AI-powered perception | Workers and AMRs share the same map of the warehouse, with live position data of each of the vehicles | Turning every movement in warehouse into shared spatial awareness that serves operators, machines, and managers alike | Equiping AGVs and other types of wheeled vehicles with multi-camera, industrial-grade Visual SLAM, providing accurate 3D positioning | Combining Visual SLAM with AI-driven 3D perception and navigation | Extending visibility to manually operated vehicles, such as forklifts, tuggers, and other types of industrial trucks | Unifying spatial awareness across fleets | Unlocking operational visibility | Ensuring every movement generates usable data | Providing foundation for smarter, data-driven decision-making | Merging manual and autonomous workflows into a single connected ecosystem | Real-time vehicle tracking | Traffic heatmaps | Spaghetti diagrams | Predictive flow analytics | Redesigning layouts | Optimizing pick paths | Streamlining material handling | Accurate vehicle tracking | Safe-speed enforcement | Pedestrian proximity alerts | Lowerung insurance claims | Ensuring regulatory compliance | Making equipment smarter, scalable, interoperable, and differentiable | Predictive maintenance | Fleet optimization | Visual AI Ecosystem connecting machines, people, processes, and data | Autonomous robotic floor cleaning | Industry 5.0 by adding people-centric approach | Visual AI to providing real-time, people-centric decision-making capabilities as part of autonomous navigation solutions | Collaborative Navigation transforming Autonomous Mobile Robots (AMRs) into mobile cobots | Visual AI confering robots the ability to understand the context of the environment, distinguishing between unobstructed and obstructed paths, categorizing the types of obstacles they encounter, and adapting their behavior dynamically in real-time | Automatically generating complete and very accurate 3D digital twin of an elevator shaft | Autonomous eTrolleys tackling last-mile problem |Autonomous product delivery at airports

#Immediate.Measures to Increase American Mineral Production

#Beijing export restrictions on rare earth minerals and permanent magnets | Foreign firms must obtain Chinese government approval to export foreign-produced rare earth magnets and semiconductor materials that contain 0.1% or more Chinese rare earth materials—or produced using Chinese mining, processing, or magnet-making technologies | Export to U.S. defense industries has been effectively banned | China accounts for some 70% of rare earth mining, 90% of processing, and 93% of magnet manufacturing | American semiconductors, jet engines, EV batteries, and military weaponry hit | China wants access to advanced semiconductor chips that U.S. has blocked it from acquiring | Rare earths are vital for chip assembly | Government export licenses are required for any rare earths used in semiconductor manufacturing or testing equipment, including sub-14-nanometer chips, or high-spec memory chips | When you take out semiconductors, magnets, or minerals, supply chain collapses | China unveiled incredibly broad export curbs on lithium and especially components of EV batteries—including battery packs, cells, cathodes, lower value anodes, as well as production equipment such as mills, furnaces, and air grinders | Minerals like magnesium, titanium, or cobalt possibly weaponized | China has a rare earths wild card

#Critical minerals in Artificial Intelligence | At the core of AI transformation lies a complex ecosystem of critical minerals, each playing a distinct role | Boron: used to alter electrical properties of silicon | Silicon: fundamental material used in most semiconductors and integrated circuits | Phosphorus: helps establish the alternating p-n junctions necessary for creating transistors and integrated circuits | Cobalt: used in metallisation processes of semiconductor manufacturing | Copper: primary conductor in integrated circuits | Gallium: used in compound semiconductors such as gallium arsenide (GaAs) and gallium nitride (GaN) | Germanium: used in high-speed integrated circuits and fibre-optic technologies | Arsenic: employed as a dopant in silicon-based semiconductors | Indium phosphide: widely used in optical communications | Palladium: used in production of multi-layer ceramic capacitors (MLCCs) | Silver: the most conductive metal used in specialised integrated circuits and circuit boards | Tungsten: serves as a key material in transistors and as a contact metal in chip interconnects | Gold: used in bonding wires, connectors, and contact pads in chip packaging | Europium: enables improved performance in lasers, LEDs, and high-frequency electronics essential to AI systems and optical networks | Yttrium: improves the efficiency and stability of materials like GaN and InP, supporting advanced applications in photonics, high-speed computing, and communications technologies

#Critical minerals in Data Storage for Artificial Intelligence | Lithium: powers batteries that support portable storage devices, SSDs, and memory-rich electronics, in data centres, lithium-ion battery arrays ensure continuous power to storage systems | Silicon: the heart of solid-state drives (SSDs) and memory chips | Manganese: next-generation memory technologies, such as resistive RAM (ReRAM) and spintronic memory, also used in lithium-ion batteries for backup power in enterprise storage systems | Praseodymium: helps improve the efficiency and durability of motors in HDDs | Neodymium: key component in neodymium-iron-boron (NdFeB) permanent magnets used in spindle motors of hard disk drives (HDDs).| Samarium: used in samarium-cobalt (SmCo) magnets, which offer exceptional thermal stability, these magnets are favoured for mission-critical and defence-grade storage systems | Gadolinium: used in advanced memory storage systems. It plays a role in magneto-optical storage technologies | Dysprosium: important for data storage applications | Platinum: used in the manufacture of high-purity glass and crucibles required in the production of memory and storage devices | Gold: used in memory chips

#Critical minerals for Optics, Imaging & Advanced Materials | Graphite: high-speed electronics, advanced sensors, and thermal management systems | Copper: short-distance data transmission in AI data centres | Germanium: a key material in thermal imaging, night-vision optics, and fibre-optic communication systems | Indium: optical communication systems | Praseodymium: specific types of lasers and optical materials | Neodymium:solid-state lasers | Holmium: specialised laser systems, particularly medical and scientific applications

#Critical minerals for Power Supply & Batteries | Lithium: portable electronics, wearables, electric vehicles | Graphite: stores lithium ions during charging process and releases them during discharge | Manganese: used in various lithium-ion battery chemistries | Cobalt: critical to the performance of premium mobile and computing devices | Nickel: crucial for electric vehicles, high-performance electronics, and energy-intensive AI systems

#Smart Lipstick application system | Automation to personal grooming for individuals with disabilities and motor limitations | Smart Lipstick device leverages artificial intelligence, computer vision, and robotics | CESAR Innovation Hub AI collabiration | Partnership with Neurobots for prototype development | Financial support of EMBRAPII (Brazilian Industrial Research and Innovation Company) | Ergonomic chin platform, similar to eye examination machines in optical clinics | User places her chin on platform, and device handles rest automatically | Activated through simple button commands | System employs a mechanical arm | Lipstick application process in approximately two minutes | Dexterous arm | Advanced computer vision | Device captures facial images to map lips and differentiate lip tones from facial colors accurately | Robotic arm applies lipstick precisely, followed by a verification photograph to confirm a correct and uniform application | Technology not only applies makeup flawlessly but empowers people with disabilities (motor disabilities or visual impairments) to experience the joy and confidence that comes with applying and wearing lipstick independently | Oromoting more autonomy during daily routines | Ability to express oneself freely, considering specific needs and preferences

#Neuroengineering solutions | Making health technology accessible to everyone | Offering cutting-edge solutions that integrate health and technology to accelerate recovery and transform lives | Smart Lipstick device to Grupo Boticario

#GenAI moving from cloud to Edge

#Tile programming | NVIDIA CUDA Tile programming | Matrix multiplication | Basis for solving systems of equations | Underpins graphics, simulations, optimization, and most of machine learning | Taking the dot product of a row of A and a column of B | Tile programming divides the output matrix into multiple tiles | Each Block is responsible for the calculation of one output tile | Kernel running on GPU | Launch code running on CPU | Standard Python function as a GPU kernel | Each matrix block run an independent instance of kernel function | Compile-time constants. cuTile will generate specialized machine code for different tile size values | Each block computes a specific tile of the output matrix | Stream: Specifies which CUDA stream the kernel executes on (for asynchronous execution and multi-stream concurrency) | Grid: Defines how many blocks to launch |.Kernel function: The GPU kernel to execute (function decorated with @ct.kernel) | Process of implementing a GPU kernel using cuTile

#Claws | Agents that act independently | Agent capable of grasping tools and pull information rather than just processing text | Claws can run continuously in background | Designed to run directly on PC | Can run shell commands | Can read and write files to reach specific objective | Work flow oriented | Self-evolving autonomy | NemoClaw open source stack | NVIDIA Nemotron | NVIDIA OpenShell runtime | NVIDIA Agent Toolkit | NemoClaw simplifies and secures AI agent deployment | NVIDIA Agent Toolkit provides full deployment stack

#Airline AI agent | Identifying booking | Understanding verbal change request | Proposing new options | Articulating fare differential | Initiating payment | Handling multiple calls in traveller preferred language | Ability to plan, book and service customised trips | Identifying opportunities across airline touchpoints like website, mobile or call centre | Supporting aircraft turnaround by monitoring maintenance, crew, re-fuelling and other processes to recommend integrated planHelping airlines to package customized offers and tailored digital experiences

#Agent Development Lifecycle (ADLC) | Build | Test | Deploy | Monitor | Improve over time

#Agentic AI | Systems for agentic AI | NVIDIA Blackwell | Agent breaks goal into many steps and keeps going until task is done | LLM calls are chained together | Each LLM call is passing growing context | Context embed tools such as code call, database search, web search | Chained LLM calls, tool call delays and growing context stress accelerated computing systems in fundamentally different ways than a single LLM call | For companies building and deploying agents at scale, it is important to understand how responsive agents are, how many can be deployed simultaneously and how much useful work AI infrastructure can deliver for every dollar and watt invested | CUDA kernels accelerate performance by overlapping communication and compute | Separating inputs processing from outputs processing helos to optimize them independently | Real coding agent trajectories: receiving a task, reading files, writing and editing code, executing commands, iterating based on results | Public code repositories across 12+ programming languages used | Platform can support simultaneously limited number of agentic tasks | There is growing demand on agentic AI at scale

#LLM | Large Language Model

#Open Model | Model whose weights have been released publicly by model creator

#Native Tokens | Tokens generated by LLM own tokenizer

#Event based vision | prophesee.ai | Combining neuromorphic sensing and bio-inspired processing to create event-based vision systems that function like eye and brain | Each pixel only reports when it senses movement | Building visual-tactile datasets for development of better learning systems in robotics | Helping robots grip and identify objects | Gesture recognition and tracking | Counting and measuring at a rate of >1,000 objects/sec | Inspecting objects at >10m/s with 100x less data to be processed | High-speed recognition applications with blur-free asynchronous event output (i.e OCR) | Measuring vibration frequencies from Hz to Khz | Understanding fine motion in scene | Understanding the finest motion dynamics hiding in ultra fast and fleeting events | Event-based sensors to track objects with low compute power | Object counting and gauging – pharmaceutical pill counting – Mechanical part counting

#Synthetic AI | Synthetic AI | AI models trained heavily on artificially generated synthetic data as well as advanced human-mimicking agents | Using synthetic approaches allows companies to bypass data shortages and navigate strict privacy regulations | It can inadvertently lead to model degradation and a perception gap in real-world performance | Strict Privacy Compliance: Eliminates exposure of personally identifiable information (PII) | Allows safe data sharing under regulations like GDPR or HIPAA | Massive Scalability: generates billions of targeted data points quickly | Developers can easily scale training sets without waiting on slow human collection | Exceptional Edge-Case training: simulates rare, high-value scenarios, useful for training autonomous vehicles or uncovering financial fraud | Lower Operational Costs: avoids expensive, bureaucratic logistics of gathering and labeling physical data | Highly Lifelike Interactions: synthetic conversational agents adapt their tone and fluidly understand human intent which drastically reduces rigid, robotic feel of standard chatbots | Training AI models recursively on synthetic data without real-world baseline updates can cause severe model degradation over time | Perception Gap": models can easily overfit and ace structured, quantitative benchmarks, yet, frequently stumbling when handling messy, unpredictable real-world scenarios | Amplified Biases: If seed dataset used to generate synthetic data contains any historical or cultural biases, generation algorithm will often inherit and heavily amplify those flaws | Lack of Genuine Innovation: synthetic data replicates existing statistical patterns and cannot discover entirely novel ideas, behaviors, or qualitative knowledge | Complex Infrastructure Demands: building accurate generative pipelines requires specialized mathematical expertise and validation servers to verify data fidelity

#Embedded Vision Summit | Incorporating computer vision and AI in products | Vision-language, large multimodal, large language and vision-language-action models | Small language models, compact VLMs, model optimization and compression, quantization, pruning, distillation and efficient edge inference | World models and physical AI systems that perceive, reason, plan and act in real-world environments, including generative and joint-embedding predictive architecture (JEPA) approaches | Deep neural networks, transformer-based networks, state-space models and neuromorphic algorithms such as spiking neural networks | Learning at edge: few-shot learning, continuous learning | 3D perception, including SLAM and scene understanding using depth sensors, radar, lidar | Sensor integration and fusion in ML-based systems—vision, depth, audio, haptic

#Think tokens in AI | Inside think tokens is AI Chain of Thought (CoT), which represents its internal reasoning process before it outputs a final answer | Reasoning contains: | Problem analysis: breaking down complex prompts into smaller, manageable parts | Fact retrieval: searching internal knowledge or planning search queries | Step-by-step logic: solving math, coding, or logic problems sequentially | Self-correction: catching mistakes, evaluating alternative approaches, and refining strategy | Safety checks: reviewing request against safety guidelines | Higher accuracy: giving AI time to think drastically improves its performance on complex tasks | Transparency: allows users to see exactly how AI arrived at a specific conclusion | Debugging: developers can look inside thoughts to find where a logic chain broke down | In AI interface like DeepSeek-R1 or OpenAI reasoning model, text between these tokens is hidden behind a collapsible Thinking Process dropdown so it does not clutter final response

#Chain of Though token | Chain of Thought token (CoT) | Any individual unit of data (word, syllable, or character) generated by Large Language Model (LLM) while it formulates its intermediate reasoning steps | Acts as model internal scratchpad | Allows midel to map out complex logic, solve multi-step problems, and self-correct before presenting a final conclusion | Autoregressive context: LLMs generate text one token at a time | Each CoT token produced serves as immediate context for the next token, building a step-by-step logic chain | CoT tokens function similarly to variables in a computer program, temporarily storing values and intermediate states required to solve broader task | Modern reasoning models allocate a specific internal thinking budget of tokens to handle complex problems, a higher number of thinking tokens usually correlates to better accuracy on difficult tasks | Visible CoT tokens are generated directly in visible text output, usually prompted by phrases like lets think step by step | Standard models use standard CoT prompting via Prompt Engineering Guide | Hidden (Internal) CoT Tokens are processed behind scenes in a native thinking phase before any text is shown to user | Advanced reasoning models separate compute stage from final response | If model must generate hundreds or thousands of intermediate tokens, time-to-response increases significantly | API providers charge for CoT tokens at standard output token rate, meaning thinking increases overall cost of query | Overthinking: models can waste tokens over-analyzing simple questions that they could have easily answered directly | Alternative frameworks like Chain of Draft (CoD) or compression tools like TokenSkip are used to dramatically minimize token footprint while keeping reasoning sharp

#NVIDIA.$500 billion initiative | Establishes independent compute financing platforms to turn AI hardware into a brand-new financial asset class | Announced via Memorandums of Understanding (MOUs) in August 2026 | NVIDIA has partnered with six of Wall Street premier asset managers | Apollo Global Management | BlackRock | Blackstone | Brookfield Asset Management | Goldman Sachs | KKR | Core objective is to treat graphics processing units (GPUs) and AI factories as income-generating infrastructure, similar to commercial real estate, toll roads, or aircraft leases | Third-Party Capital Mobilization | Wall Street firms will source, vet, and individually underwrite loan proposals for hyperscalers, frontier labs (like OpenAI and Anthropic), and enterprises | GPUs as Loan Collateral: borrowers secure massive loans using NVIDIA hardware itself as collateral, functioning on premise that compute has clear intrinsic resale and rental value | NVIDIA Financial Backstop: NVIDIA provides residual support, promising to backstop up to 25% (or $125 billion) of individual deals to stabilize hardware value if a borrower defaults | Secondary Liquidity Ecosystem: If a client defaults, NVIDIA and its partners plan to quickly re-rent or relocate affected chips to other waitlisted data centers, protecting enders from total capital loss | Wall Street financial engineering introduces massive benefits for NVIDIA corporate ecosystem and financial metrics | Handing credit analysis and capital pool over to independent institutional giants validates genuine market demand | Unlocks kong-duration revenue share: beyond selling silicon upfront, NVIDIA could capture up to a 35% revenue share above breakeven from these platforms, potentially adding a 10%+ upside to FY2029 earnings per share | Secures CUDA ecosystem: by subsidizing and simplifying financing hurdle for startups and enterprises, NVIDIA locks customers deeper into its proprietary CUDA software stack, keeping competitors out | BlackRoc CEO Larry Fink likened this initiative to 1970s creation of mortgage-backed securities, calling it the next era of financial engineering | If underlying economic demand for AI tokens and services keeps pace, this structure ensures NVIDIA remains undisputed gatekeeper of global infrastructure | Bringing independent long-term institutional capital to infrastructure market demand is genuine | 35% revenue share above breakeven could provide more than 10% upside Nvidia fiscal 2029 earning

#Unitree IPO in Shanghai | Unitree Robotics became the first humanoid robot maker listed on A-share market in Shanghai | The first humanoid company to go public in mainland China | Chinese robotics giant Unitree soars in stock market debut | Unitree Robotics stock soars 460% in Shanghai IPO debut | Shares of Unitree surged nearly 630% in China, before closing up 460% | Company raised $900 million in its debut | Strategic investors include Chinese AI startup DeepSeek, a group associated with tech giant Tencent, and several state-owned utility companies | Retail traders were 5,000x oversubscribed | China humanoid market is predicted to grow from $2 billion 2026 to $15 billion by 2030 | IPO price of 150.80 yuan with stock closing at 845 yuan represented a 460 per cent gain | Unitree move toward capital market sends important signal: humanoid robotics and embodied AI are moving beyond technology development, competition-based validation and product iteration toward industrialization, scalability and broader recognition from capital market | Hangzhou-based company offered ca. 40.45 million shares at 150.8 yuan each, representing a price-to-earnings ratio of 219.23 | Its cumulative quadruped robot shipments exceeded 33,000 units, with a global market share of nearly 60 percent | Unitree specializes in quadruped and humanoid robots | Unitree has fully self-developed core components, including motors, reducers, controllers, and LiDAR | Company posted revenue of about 1.15 billion yuan in the first half of 2026, up 48.54 percent year on year | Funds raised will be put toward intelligent robot model development, robot hardware R&D, new product development and manufacturing base construction | Business moves from robot manufacturing toward building a broader ecosystem for high-performance general-purpose robots | Unitree founder Wang Xingxing was quoted by Shanghai Securities News | Unitree unveiled its new humanoid robot Superman | Global humanoid robot shipments are projected to exceed 510,000 units by 2030

#Nvidia and AI Boom | No sign of slowing | Forecasts even more AI spending | AI has reached its inflection point | AI is doing useful work | Tokens are productive and profitable | Compute is revenue | Physical AI is coming online | Nvidia newer CPU business, specifically built to power agentic AI, continues to ride strong momentum | 26 Aug 2026 reported quarterly revenue of $96.2 billion | Nvidia is considered the bellwether for AI spending frenzy, with its results largely determined by mega purchases from AI giants such as OpenAI, Amazon, Microsoft and xAI | Revenue for quarter ending July 26 rose 106% from a year earlier and 18% from previous three months, comfortably above $92 billion analysts had forecast | Net profit reached $59.7 billion, up 126% from a year earlier | Adjusted earnings per share $2.22, up 120% year over year | $7.8 billion in gains from Nvidia portfolio of stakes in AI companies | Amazon, Microsoft, Google-parent Alphabet and Meta are on track to spend roughly $800 billion on data centers and AI infrastructure 2026 | Analysts expecting industry-wide spending to top $1 trillion | Nvidia bankrolling AI expansion through investments in startups | Nvidia faced questions about circular financing through investments in its own startup clients, helping them pay for Nvidia products | Major computing platform shift going on

#Geospatial AI | Collection problem largely solved with point clouds and oriented images | Challenge to deciding which points are ground and which are vegetation, finding kerb line, checking whether survey actually met tolerance, and turning all of it into something designers or asset managers can use | Gap is where geospatial AI is being applied, and it is quietly changing what mapping technology means in practice | Machine learning models are trained to recognise patterns in spatial data: classifying a point cloud, extracting features from imagery, flagging measurements that look wrong | Separating ground from vegetation, buildings, poles and wires | Road markings, kerbs, signs, manholes and facade lines can be identified in imagery or point clouds and turned into vectors | Quality control | Volume calculation | Reality capture, practice of recording whole scene rather than chosen set of points, has become normal work rather than specialist service | CHC Navigation integrated hardware and software workflows across GNSS, IMU, vision and LiDAR are designed so that positioning, imagery and point clouds arrive already aligned and time-stamped, which is condition any automated interpretation depends on | Classification model can tell that a set of points is a kerb but it cannot tell you where that kerb is | Position comes from GNSS, from inertial measurement, and from way those are fused into 5trajectory, and any error there propagates through everything the model produces afterwards | Accuracy questions have not gone away: Multipath in urban canyon, short GNSS interruption under bridge, correction service that drops for thirty seconds: each one puts a small distortion into trajectory | Automated classification that comes with confidence measure, and clear way to see which areas model was unsure about

#Precision Grading | Cutting and filling ground to a design surface within a tight tolerance, often a few centimetres or less, so that pads, roads, drainage and foundations sit exactly where plan calls for them | GPS grade control system | 3D design surface | GNSS RTK positioning | Machine sensors | In-cab grade guidance for accurate grading results | Automatic blade control keeps dozer blade on design surface as it pushes material | Grade control on a motor grader holds moldboard to design slope and elevation so finished surface is smooth and consistent in a single pass | On excavator, grade control guides bucket to correct depth and slope for foundations, trenches and drainage, so operator can dig to grade without a grade checker in trench | 3D excavator machine control system brings design surface and bucket position into cab for exactly this kind of work | Machine keeps working instead of pausing for grade checks| Fewer people on foot near working machines is a safety gain in its own right | Precise grading also reduces two most expensive kinds of waste on earthworks site: rework and over-excavation | Grade control systems install on machines already in fleet | With real time kinematic corrections from a base station or a correction network, a grade control system positions the cutting edge to centimetre-level accuracy