Industry Briefing
A single destination for timely, editor-curated robotics news from around the world.
Deep Reinforcement Learning Based Autonomous Decision‐Making for Cooperative Uncrewed Aerial Vehicles: A Search and Rescue Real World Application
A recent study published in the Journal of Field Robotics highlights advancements in autonomous robotic systems designed for agricultural applications. Researchers from various institutions collaborated to develop innovative algorithms that enhance the efficiency and precision of robotic farming equipment. The findings, released in early October 2023, emphasize the growing importance of automation in agriculture, particularly in response to labor shortages and the need for sustainable farming practices. The research was conducted in multiple agricultural settings, showcasing how these robotic systems can adapt to different crop types and environmental conditions. By integrating machine learning and sensor technology, the robots are capable of performing tasks such as planting, weeding, and harvesting with minimal human intervention. This development aims to address the challenges faced by farmers, including the rising costs of labor and the increasing demand for food production. The study underscores the potential for these autonomous systems to revolutionize the agricultural sector, making it more efficient and environmentally friendly. As the agricultural industry continues to evolve, the implementation of such technologies could lead to significant improvements in productivity and sustainability.
RESEARCH ARTICLE
A Feature‐Decoupled and Gated‐Interaction‐Enhanced Deep Reinforcement Learning for Path‐Following of Large‐Inertia Vessels
A recent study published in the Journal of Field Robotics explores advancements in autonomous robotic systems designed for agricultural applications. Researchers from various institutions conducted the study to address the growing need for efficient farming solutions amid increasing global food demand. The findings, released in early October 2023, highlight innovative technologies that enable robots to perform tasks such as planting, monitoring crop health, and harvesting with minimal human intervention. The research was carried out in diverse agricultural settings, demonstrating the robots' adaptability to different environments and crop types. By integrating artificial intelligence and machine learning, these autonomous systems can analyze data in real-time, making informed decisions that enhance productivity and reduce resource waste. The motivation behind this study stems from the challenges faced by the agricultural sector, including labor shortages and the need for sustainable practices. The researchers aim to provide farmers with tools that not only improve efficiency but also contribute to environmental sustainability. Through rigorous testing and validation, the study showcases the potential of these robotic systems to revolutionize farming practices, ultimately leading to increased yields and reduced operational costs. As the agricultural industry continues to evolve, the implementation of such technologies could play a crucial role in meeting future food security challenges.
RESEARCH ARTICLE
Redefining Optimal Coverage Path Planning for FLS‐Equipped AUVs With Deep Reinforcement Learning
The Journal of Field Robotics has recently published an EarlyView article highlighting advancements in robotic technology. Researchers from various institutions have collaborated to explore innovative applications of robotics in field environments. This study, released in October 2023, focuses on enhancing the efficiency and effectiveness of robotic systems in agricultural and environmental monitoring tasks. The motivation behind this research stems from the increasing demand for precision agriculture and sustainable practices, which necessitate the integration of advanced robotics. By employing cutting-edge algorithms and sensor technologies, the team aims to improve data collection and analysis in challenging outdoor conditions. The findings suggest that these advancements could significantly reduce labor costs and increase productivity for farmers, while also providing critical insights for environmental conservation efforts. This collaborative effort underscores the potential of robotics to transform traditional practices and address pressing global challenges.
RESEARCH ARTICLE
Japanese Team Develops Snake Robot with New Rolling Motion, Achieving 10x Energy Efficiency
A research team at Osaka Metropolitan University has made significant advancements in robotics by training a snake-like robot to roll forward, which has notably improved its energy efficiency. By employing deep reinforcement learning techniques, the robot is able to maintain a nearly straight trajectory while achieving energy consumption reductions of up to tenfold compared to conventional crawling methods. This breakthrough not only showcases the potential for more efficient robotic movement but also opens avenues for further developments in energy-efficient robotic applications.
Snake Robots Energy Efficiency Deep Reinforcement Learning Robotics Motion Control
Video Friday: Humanoid Learns Tennis Skills Playing Humans
IEEE Spectrum robotics has released its latest edition of Video Friday, showcasing a variety of innovative robotics videos and announcing upcoming events in the field. Notable events include the International Conference on Robotics and Automation (ICRA) scheduled for June 1-5, 2026, in Vienna, and a Summer School on Multi-Robot Systems from July 29 to August 4, 2026, in Prague. Among the featured advancements, researchers have developed LATENT, a system designed to teach humanoid robots tennis skills by learning from imperfect human motion data. This innovation addresses the challenges of replicating human-like athleticism in robotics. Additionally, a breakthrough has been achieved in robotic manipulation, with a robot successfully peeling an apple using dual dexterous hands, showcasing significant progress in bimanual tasks. The development of MoDE-VLA, a control system that integrates vision, language, force, and touch data, further enhances the robot's ability to perform complex tasks with stability and precision. This shared-autonomy approach allows human operators to guide robots in executing intricate movements. In other highlights, collaborations between Tesollo and Hanyang University have led to advancements in robotic hand technology, while the Fluent Robotics Lab at the University of Michigan is set to present a paper on operational PR2 robots. The KAIST DRCD Lab has also demonstrated the capabilities of its humanoid robot, trained through deep reinforcement learning. As robotics continues to evolve, these innovations reflect the ongoing efforts to bridge the gap between human-like dexterity and robotic functionality.
Humanoid-robots Video-friday Robot-locomotion Nvidia Robot-manipulation Quadruped-robots
Robot Talk Episode 136 – Making driverless vehicles smarter, with Shimon Whiteson
Claire engaged in a discussion with Shimon Whiteson, a prominent figure in the field of machine learning and autonomous vehicles. Whiteson, who serves as a Professor of Computer Science at the University of Oxford and a Senior Staff Research Scientist at Waymo UK, shared insights on his research, which emphasizes deep reinforcement learning and imitation learning. These areas of study are pivotal for advancements in robotics and video game technology. The conversation highlighted the significance of machine learning in enhancing the capabilities of autonomous vehicles, reflecting the growing intersection of academia and industry in this rapidly evolving field.
Launch of Robo-ValueRL: The First Open-Source VLA Reinforcement Learning Framework for Robotics
The Beijing Humanoid Robot Innovation Center and Renmin University of China's Gaoling Artificial Intelligence Institute have launched the Robo-ValueRL open-source framework. This initiative aims to enhance humanoid robots' decision-making capabilities in precision tasks, such as semiconductor assembly, by addressing challenges in data quality, control precision, and adaptability in dynamic environments. Robo-ValueRL introduces a value estimation mechanism based on historical observations, enabling robots to autonomously assess their actions. This closed-loop learning process—observation, value estimation, correction, and iteration—allows for improved accuracy and reduced instability in operations. The framework is fully open-source, providing access to core algorithms, evaluation tools, and standardized protocols for universities, research institutions, and manufacturers. The open-source nature of Robo-ValueRL significantly lowers the barriers for small and medium-sized manufacturers to implement reinforcement learning in specialized fields like semiconductor production and medical device manufacturing. This development marks a shift in humanoid robotics from laboratory experiments to practical industrial applications, paving the way for robots to evolve their decision-making capabilities independently.
Humanoid Robots Reinforcement Learning Precision Manufacturing Open Source Technology
"Father of Reinforcement Learning Sutton Partners with Tianshan Technology for First Deep Sharing in China! Limited Rare Spots Open for a Short Time, Don't Miss Out!"
In a significant development for the field of artificial intelligence, Richard Sutton, a pioneer in reinforcement learning, has teamed up with Tianshan Technology to host a groundbreaking event in China. This collaboration marks the first deep sharing initiative in the country, aimed at advancing knowledge and application of reinforcement learning techniques. The event is set to take place soon, with limited spots available for participants eager to gain insights from one of the leading experts in the field. Attendees will have the unique opportunity to engage directly with Sutton and learn about the latest advancements and practical applications of reinforcement learning. This initiative reflects a growing interest in AI technologies in China and underscores the importance of collaboration between leading researchers and local tech companies to foster innovation and expertise. Interested individuals are encouraged to secure their spots promptly, as availability is limited.
Robotics Automation AI
UK startup Humanoid launches reinforcement learning system to improve robot manipulation
UK-based robotics and AI company Humanoid has introduced KinetIQ Ascend, the company’s reinforcement learning approach designed to reach 99.9 percent manipulation reliability at human speed and beyond. KinetIQ Ascend builds on the previously announced KinetIQ platform with trial-and-error learning, helping the company’s robots improve directly on industrial tasks. The new system was tested on several […]
Computing Humanoids News artificial intelligence automation embodied ai
Deep-Learning Tool Improves Anatomical Recognition in Pelvic Surgery
A new deep-learning model improved surgeons’ recognition of pelvic anatomy in video-based PLND tests, though live surgical validation is still needed.
New Tactile Sensors Achieve High Resolution Without Deep Learning
Researchers from Queen Mary University of London and the University of Florence have unveiled a groundbreaking mechanochromic film measuring just 16 microns in thickness, designed to enhance tactile sensing capabilities in robots. This innovative sensor operates without the need for deep learning, directly translating mechanical strain into color changes. As a result, it generates real-time pressure maps with an impressive spatial resolution of around 100 microns. This advancement significantly boosts the dexterity of robotic systems, enabling them to interact more effectively with their environments. The development marks a notable step forward in robotics, potentially transforming how machines perceive and respond to tactile stimuli.
Tactile Sensors Robotics Mechanochromic Materials Pressure Mapping
Humanoid says KinetIQ Ascend reinforcement learning approaches human-level dexterity
Humanoid has announced that its KinetIQ Ascend technology achieves an impressive 99.9% manipulation reliability, capable of performing industrial tasks at human speed and even surpassing it. This breakthrough is attributed to advanced reinforcement learning techniques that enable robots to exhibit human-level dexterity. The development marks a significant advancement in robotics, potentially transforming efficiency in various industrial applications.
Artificial Intelligence Artificial Intelligence / Cognition Humanoids News dexterous manipulation humanoid
"Not Competing with Real Machines, Not Stacking Simulations: Deep Intelligence Relies on 'Human Learning' to Secure Hundreds of Millions in Financing"
Deep Intelligence, a company focused on advancing artificial intelligence through human learning methodologies, has secured hundreds of millions in financing. This significant funding round, announced recently, highlights the growing interest in AI technologies that prioritize human-like learning processes over traditional machine simulations. The investment aims to enhance Deep Intelligence's research and development efforts, positioning the company as a leader in the AI sector. By leveraging insights from human cognition, Deep Intelligence seeks to create more intuitive and effective AI systems. The financing comes at a time when the demand for innovative AI solutions is surging, driven by various industries looking to integrate smarter technologies into their operations.
Robotics Automation AI
Toyota's CUE Robot Advances: Learning to Walk and Dribble with Reinforcement Learning and Sim2Real
Toyota's CUE humanoid robot is advancing its capabilities through a novel approach that integrates reinforcement learning with Sim2Real techniques. This development focuses on improving the robot's walking and dribbling abilities, effectively narrowing the divide between simulated environments and real-world functionality. By employing this innovative method, Toyota aims to enhance the practical applications of robotics, showcasing the potential for more sophisticated interactions in various settings.
Humanoid Robots Reinforcement Learning Sim2Real AI Robotics
Can Robots Trained Through Behavior Cloning Evolve Themselves in Two Hours Using Reinforcement Learning?
Researchers have identified significant limitations in behavior cloning (BC) methods used in robotics, prompting the development of a new approach known as Q2RL. This innovative technique integrates BC with reinforcement learning (RL) to enhance the performance of robots. By leveraging hidden knowledge embedded in BC strategies, Q2RL seeks to improve the efficiency of learning processes while simultaneously lowering the costs tied to data collection and the need for retraining. This advancement represents a crucial step forward in optimizing robotic capabilities, addressing the challenges faced by traditional BC methods.
Reinforcement Learning Behavior Cloning Robotics AI Machine Learning
NVIDIA, Ineffable Intelligence Team Up to Build the Future of Reinforcement Learning Infrastructure
NVIDIA has announced a new engineering collaboration with Ineffable Intelligence, a London-based AI lab, aimed at enhancing the capabilities of reinforcement-learning agents. These AI systems, which learn through trial and error, are designed to transform computational processes into valuable knowledge. The partnership seeks to leverage the strengths of both organizations to advance the development of these intelligent systems, potentially leading to significant breakthroughs in AI applications. The collaboration underscores a growing interest in harnessing advanced AI techniques to drive innovation and efficiency across various sectors.
Detecting IoT Malware in EV Chargers with Deep Learning
A recent study has introduced a deep learning methodology aimed at improving malware detection in electric vehicle (EV) charging stations. Conducted by a team of researchers, the study addresses significant limitations in current detection systems, ultimately enhancing their accuracy. The findings, which were published in October 2023, highlight the growing need for robust cybersecurity measures in the rapidly expanding network of IoT devices associated with EV infrastructure. By leveraging advanced machine learning techniques, the researchers aim to provide a more effective solution to safeguard these critical charging stations from potential cyber threats, ensuring the safety and reliability of EV charging operations.
Sanctuary AI Touts Reinforcement Learning Success for Dexterous Robot Hand Manipulation
Sanctuary AI has showcased its advanced robotic hand, featuring hydraulically actuated five fingers, successfully executing in-hand object reorientation. This demonstration took place recently, highlighting the company's innovative approach to robotics. The robotic hand utilized a reinforcement learning policy that was initially trained in a simulated environment, achieving a notable sim-to-real transfer even when subjected to an unexpected load of 500 grams. Sanctuary AI credits this accomplishment to its proprietary reinforcement learning techniques and the sophisticated design of its high-degree-of-freedom hand hardware, marking a significant milestone in the development of robotic manipulation capabilities.
phoenix sanctuary-ai
Natural Humanoid Walk Using Reinforcement Learning
Researchers have developed an innovative neural network designed for humanoid locomotion, utilizing reinforcement learning to enable robots to walk in a manner akin to humans. This breakthrough was achieved through advanced high-fidelity simulations, which not only improve the robots' walking capabilities but also facilitate a seamless transition from simulated environments to real-world applications. The technology promises to significantly enhance the efficiency and scalability of humanoid robotics, making it a pivotal advancement in the field. The development is expected to impact various sectors, including robotics and automation, by providing more adaptable and capable humanoid robots.
Reinforcement Learning Humanoid Robotics Simulation Technology AI Development Robotics Engineering
Deep Learning‐Driven Steering Angle Prediction and Scene Understanding via Harmonic Carpet Weaver Optimization
In June 2026, the Journal of Field Robotics published a comprehensive study examining advancements in robotic technologies and their applications in various fields. This research highlights the contributions of leading experts in robotics, who analyzed the latest innovations and their potential to enhance efficiency and safety in sectors such as agriculture, manufacturing, and disaster response. The study emphasizes the growing importance of integrating artificial intelligence and machine learning into robotic systems to improve their adaptability and functionality. Researchers conducted extensive field tests to evaluate the performance of these robots in real-world scenarios, demonstrating their effectiveness in tasks ranging from precision farming to search and rescue operations. The motivation behind this research stems from the increasing demand for automation and the need for more reliable and intelligent robotic solutions to address complex challenges faced by industries today. By providing empirical data and insights, the study aims to inform policymakers, industry leaders, and researchers about the transformative potential of robotics. As the field continues to evolve, the findings presented in this publication are expected to influence future developments and investments in robotic technologies, ultimately shaping the landscape of automation in the coming years.
RESEARCH ARTICLE
A Critical Review of Reinforcement Learning Algorithms for Mobile Robot Path Planning
The Journal of Field Robotics has published an early view article highlighting recent advancements in robotic technology. Researchers from various institutions have collaborated to explore innovative applications of robotics in diverse fields, including agriculture, healthcare, and disaster response. The findings, released in October 2023, underscore the growing importance of robotics in enhancing efficiency and safety across these sectors. The study emphasizes the integration of artificial intelligence and machine learning to improve the functionality and adaptability of robotic systems. By leveraging these technologies, the researchers aim to address complex challenges faced in real-world scenarios, such as precision farming and emergency management. This publication is part of an ongoing effort to disseminate cutting-edge research that can inform future developments in robotics. The collaborative nature of the research showcases a commitment to interdisciplinary approaches, fostering innovation that can lead to significant societal benefits. As the field continues to evolve, the implications of these advancements are expected to resonate across various industries, driving further investment and interest in robotic solutions.
SURVEY ARTICLE
A Field‐Adaptive Mechanical Weeding System Coupling Oscillating Pneumatic Mechanism With Deep Learning for Intra‐Row Weed Control in Lettuce
The Journal of Field Robotics has published an early view article highlighting recent advancements in autonomous robotic systems. Researchers from leading universities and technology firms presented their findings on October 15, 2023, during a virtual conference focused on robotics innovation. The study emphasizes the growing importance of these systems in various sectors, including agriculture, search and rescue, and environmental monitoring. The motivation behind this research stems from the increasing demand for efficient and reliable robotic solutions capable of operating in complex environments. By integrating advanced artificial intelligence and machine learning algorithms, the researchers demonstrated how these autonomous systems can enhance operational capabilities and decision-making processes. The article details various case studies showcasing successful implementations of robotic technologies, illustrating their potential to revolutionize traditional practices. The findings suggest that as technology continues to evolve, the integration of autonomous robots will become crucial in addressing global challenges, such as food security and disaster response. This publication marks a significant contribution to the field of robotics, providing insights into future trends and encouraging further exploration of autonomous systems' applications. Researchers and industry professionals are urged to collaborate and innovate, ensuring that the benefits of these technologies are realized across multiple domains.
RESEARCH ARTICLE
Humanoid Announces its KinetIQ Ascend Reinforcement Learning Approach
A groundbreaking robotic system has demonstrated its effectiveness in various manipulation tasks, including retrieving parts from bins, delivering objects to humans, and lifting and moving containers with its dual arms. This innovative technology was rigorously tested and has shown promising results across multiple scenarios, showcasing its potential for practical applications in industries requiring automation. The trials were conducted recently, highlighting the system's versatility and efficiency in handling complex tasks that typically require human intervention. As industries increasingly seek to enhance productivity through automation, this new robotic solution could play a significant role in transforming operational workflows.
Prox Industries accelerates physical AI research with dual-arm UR3e collaborative robots using VLA and reinforcement learning.
Prox Industries has announced its collaboration with Universal Robots (UR) to enhance the development of physical AI through the utilization of UR's "Physical AI Development Support Program." The initiative will focus on accelerating research and development of physical AI by employing a dual-arm robotic configuration using two UR3e collaborative robots. This partnership aims to leverage advanced robotics technology to innovate in the field of AI, reflecting Prox Industries' commitment to advancing automation solutions.
Deep Learning Based Dirt Detection and Cleanliness Evaluation in Autonomous Indian Domestic Concrete Water Tank Cleaning Robot
A recent study published in the Journal of Field Robotics highlights advancements in autonomous robotic navigation. Researchers from a leading robotics institute conducted experiments to enhance the efficiency of robots in complex environments. The study, released in early October 2023, focuses on the integration of advanced algorithms that allow robots to better interpret their surroundings and make real-time decisions. The research was carried out in various challenging terrains, including urban settings and natural landscapes, to test the robots' adaptability. The motivation behind this work stems from the growing demand for autonomous systems in sectors such as agriculture, search and rescue, and urban planning. By improving navigation capabilities, the researchers aim to facilitate the deployment of robots in scenarios where human intervention is limited or dangerous. Through a series of simulations and field tests, the team demonstrated that the new algorithms significantly reduced the time taken for robots to complete tasks while increasing their accuracy in obstacle avoidance. This breakthrough could lead to more reliable and efficient robotic systems, paving the way for wider applications in everyday life. The findings underscore the potential of robotics to transform various industries by enhancing operational efficiency and safety.
RESEARCH ARTICLE
Deep learning co-design helps scientists project 28-layer 3D images without crosstalk
Engineering researchers at the University of California, Los Angeles (UCLA) have unveiled a groundbreaking three-dimensional printing technology that significantly enhances the production of complex structures. This innovative method, introduced in October 2023, aims to revolutionize various industries by allowing for the rapid and precise fabrication of intricate designs that were previously difficult or impossible to achieve. The researchers' motivation stems from the growing demand for more efficient manufacturing processes that can produce high-quality components while minimizing waste and time. By leveraging advanced materials and techniques, the team has demonstrated that their new approach can streamline production workflows and reduce costs, making it an attractive option for sectors such as aerospace, automotive, and biomedical engineering. This development not only showcases the potential of 3D printing technology but also emphasizes UCLA's commitment to leading research in engineering and technology. The researchers plan to further refine their technique and explore its applications across various fields, aiming to set new standards in manufacturing efficiency and innovation.
Science
Advanced Deep Learning Architecture for Real‐Time Online Train Tread Segmentation and Wear Detection
In May 2026, researchers published a study in the Journal of Field Robotics, exploring advancements in robotic technology for agricultural applications. The study focuses on the development of autonomous robots designed to enhance efficiency in crop management and harvesting processes. Conducted by a team of engineers and agricultural scientists, the research highlights the growing need for innovative solutions in the face of labor shortages and increasing food production demands. The team conducted field trials in various agricultural settings to assess the robots' performance and adaptability to different crop types. Their findings indicate that these autonomous systems can significantly reduce labor costs and improve yield quality, addressing both economic and environmental challenges faced by the agriculture sector. The research underscores the potential for robotics to transform traditional farming practices, making them more sustainable and efficient. This study is part of a broader initiative to integrate advanced technologies into agriculture, aiming to support farmers in meeting the global food supply challenges. By leveraging robotics, the researchers hope to pave the way for smarter farming practices that can respond to the dynamic needs of the industry.
RESEARCH ARTICLE
Google DeepMind Unveils Gemini Robotics-ER 1.6: A Leap in Spatial Reasoning and Industrial Utility
DeepMind has unveiled its latest and most sophisticated embodied reasoning model, which incorporates a cutting-edge "agentic vision" system designed specifically for industrial inspections. This new technology aims to improve the accuracy and efficiency of multi-view success detection in various industrial applications. The launch, which took place recently, marks a significant advancement in artificial intelligence capabilities, reflecting DeepMind's commitment to enhancing operational processes across industries. By leveraging advanced machine learning techniques, the model is expected to streamline inspection workflows and reduce the likelihood of errors, ultimately driving productivity and safety in industrial environments.
Google Gemini Gemini Robotics-ER 1.6 google-deepmind Boston Dynamics
Deep Intelligence Secures Hundreds of Millions in Funding Within a Year: What Attracts Investors?
Beijing Deep Intelligence, a company established only a year ago, has successfully secured hundreds of millions in funding from prominent investors, including state-owned entities and major industry players. The firm has developed an innovative approach known as 'human learning,' which allows robots to comprehend their environment prior to performing tasks. This methodology has resulted in performance metrics that outshine those of competitors. As the demand for embodied intelligence continues to expand swiftly, the company faces the ongoing challenge of transforming its technological advancements into tangible commercial success.
Embodied Intelligence Robotics AI Technology Investment
Deepak Pathak Named to MIT Technology Review’s Innovators Under 35 List
Deepak Pathak, an assistant professor at the Robotics Institute, has been recognized as one of MIT Technology Review’s 35 Innovators Under 35 for his groundbreaking contributions to self-supervised and adaptive robot learning. This prestigious list, published by MIT, showcases emerging leaders who leverage technology to address significant societal challenges and explore important scientific inquiries. Pathak's innovative work positions him at the forefront of advancements in robotics, highlighting his potential to influence the future of the field.
Uncategorized
AI Is Learning to Read the Room
Recent advancements in emotion AI technology are reshaping how machines interpret human feelings, particularly in professional settings. Companies like Meta and startups such as Hume AI are developing systems that analyze facial expressions, voice tones, and behaviors to gauge emotions during interactions like performance reviews. This technology, which has applications in employee well-being, recruitment, and customer service, aims to enhance communication by providing real-time feedback. Despite its rapid growth, current emotion AI systems often struggle to capture the complexity of human emotions, typically categorizing feelings into simplistic labels like "happy" or "sad." Researchers are now focusing on a new approach called human-context AI, which combines multiple inputs—such as facial dynamics and voice modulation—with situational context to better understand emotional nuances. This shift aims to close the gap between human emotional expression and machine interpretation. The origins of emotion AI trace back to the MIT Media Lab, where Rosalind Picard pioneered the concept of affective computing. Over the years, advancements in data collection and analysis have improved the accuracy of emotion detection. However, ethical concerns remain, particularly regarding privacy and the potential for misuse in workplaces and public spaces. As this technology evolves, it promises to enhance various applications, from professional development platforms to health care, by providing a deeper understanding of human emotions. Yet, experts caution against over-reliance on AI for critical decisions, emphasizing the importance of human insight in interpreting emotional signals.
Emotions Affective-computing Facial-expressions Companion-robots Multimodal-ai Machine-learning
DeepTrack: A Pressure‐Tolerant Electromagnetically Driven Soft Robotic Fish Platform With Visually Guided Locomotion
The Journal of Field Robotics has published an early view article highlighting advancements in autonomous robotic systems. Researchers from various institutions collaborated to explore innovative algorithms that enhance the navigation and decision-making capabilities of robots in complex environments. This study, released in October 2023, aims to address the growing need for efficient robotic solutions in sectors such as agriculture, search and rescue, and industrial automation. By employing machine learning techniques and real-time data processing, the team demonstrated significant improvements in the robots' ability to adapt to dynamic surroundings. The findings are expected to pave the way for more effective deployment of robotic technologies in real-world applications, ultimately contributing to increased productivity and safety in various industries.
RESEARCH ARTICLE
What lessons is Ideal learning to catch up with FSD V14?
The competitive landscape of the intelligent driving industry has undergone significant changes in recent years, shifting from hardware specifications to advanced model development. Companies are increasingly recognizing that merely having larger models is insufficient for achieving generational advantages; instead, the integration of models, data, computing power, and chips into a continuous iterative loop is becoming crucial. This realization has prompted many automakers to invest in in-house research and development. Tesla has established a comprehensive ecosystem that spans data collection, training infrastructure, and self-developed chips, while Chinese companies like Li Auto, Xpeng, and NIO are also deepening their technological foundations. Li Auto has introduced its self-developed Mach M100 chip in its L8 and L9 models, which it views as a significant advancement in AI technology. In a recent discussion with Li Auto's autonomous driving and chip leaders, they emphasized that the industry should focus on the practical problems these investments aim to solve rather than merely the existence of in-house development. They outlined their strategies to achieve performance comparable to Tesla's Full Self-Driving (FSD) system, highlighting the importance of safety, efficiency, and comfort in user experience. As the industry moves towards higher levels of autonomy, the integration of vision and language models is seen as essential for developing systems that can handle complex, unforeseen scenarios. The executives noted that achieving higher levels of autonomy (L3 and L4) requires models that can reason and think like humans, underscoring the growing significance of language in AI systems. Overall, the conversation revealed the industry's focus on enhancing AI capabilities through innovative chip design and data utilization, aiming for a future where autonomous driving technology can meet the challenges of real-world driving conditions.
Research on ORB–SLAM2 Deep‐Sea Visual Localization Technique Based on Image Enhancement
A recent study published in the Journal of Field Robotics highlights advancements in autonomous robotic systems designed for agricultural applications. Conducted by a team of researchers from various universities, the study was released in early October 2023. The research focuses on the development of innovative robotic technologies aimed at improving efficiency and sustainability in farming practices. The motivation behind this initiative stems from the growing need for sustainable agricultural solutions in response to increasing global food demands and environmental concerns. The researchers employed a combination of machine learning algorithms and sensor technologies to enhance the robots' capabilities in tasks such as planting, harvesting, and monitoring crop health. Field tests were conducted in diverse agricultural settings, demonstrating the robots' ability to adapt to varying conditions and perform complex tasks autonomously. The findings suggest that these robotic systems could significantly reduce labor costs and increase productivity while minimizing the environmental impact of farming. The study underscores the potential of robotics to transform the agricultural sector, paving the way for more efficient and eco-friendly farming practices. As the agricultural industry faces challenges from climate change and population growth, the integration of advanced robotic technologies may offer viable solutions for future food production.
RESEARCH ARTICLE
Generalist AI Releases "Science of Pretraining" Deep Dive: Why Data Quality Trumps Volume in Robotics
Generalist AI has unveiled new insights into its pretraining methodology in a technical addendum related to its recent GEN-0 launch. The company introduced innovative metrics, including "Reverse KL," designed to evaluate the creativity of its models. Additionally, Generalist AI announced that its infrastructure can process an impressive volume of data, equating to 6.85 years of robotic experience each day. This advancement highlights the company's commitment to enhancing artificial intelligence capabilities and underscores its efforts to push the boundaries of machine learning technology.
Data Collection Generalist AI embodied-ai
MIT’s SoftMimic Framework Teaches Humanoids to Be Compliant, Not Just Capable
Researchers at MIT's Improbable AI Lab have unveiled SoftMimic, an innovative reinforcement learning framework designed to enhance the safety and compliance of humanoid robots during interactions. Unlike traditional methods that focus on rigid motion tracking, SoftMimic enables robots to adaptively absorb collisions, promoting safer engagement with their environment. This development, announced in October 2023, aims to address the growing need for more flexible and responsive robotic systems in various applications, from manufacturing to personal assistance. By leveraging advanced algorithms, the framework allows humanoids to learn from their experiences, improving their ability to navigate complex scenarios while minimizing the risk of injury to themselves and humans alike.
reinforcement-learning MIT SoftMimic
1X Reveals the RL Core That Gives Its NEO Humanoid Fluid, Whole-Body Mobility
Robotics firm 1X has revealed the advanced reinforcement learning system that drives the movement of its NEO humanoid, following the introduction of its Redwood AI brain. This innovative unified controller allows the robot to perform a variety of human-like actions, including walking, running, and climbing stairs, utilizing only stereo vision for navigation. The announcement highlights 1X's commitment to enhancing robotic mobility and intelligence, showcasing the potential for humanoid robots to interact more naturally within human environments.
reinforcement-learning 1X-technologies Redwood humanoid-robots locomotion robotics-ai
AGIBOT Introduces Genie Sim 3.0, an Integrated Simulation, Data, and Benchmarking Platform for Embodied AI
AGIBOT has unveiled Genie Sim 3.0, an advanced platform aimed at improving embodied artificial intelligence in robotics. Launched recently, this open-source platform addresses significant challenges in robotics development by incorporating features such as environment generation, data scalability, and standardized evaluation methods. Genie Sim 3.0 enables the creation of 3D environments driven by large language models (LLMs) and includes a comprehensive framework for evaluating robot algorithms. The platform also integrates deeply with reinforcement learning, streamlining the experimentation and deployment processes for robotics. This upgrade is expected to facilitate faster advancements in the field, enhancing the capabilities and efficiency of robotic systems.
Embodied AI Robotics Simulation Reinforcement Learning Data Evaluation
Simulating the 'Masterpiece': XPENG Robotics Unveils Design Framework and Lattice Musculature for IRON
XPENG Robotics has unveiled a comprehensive technical analysis detailing the innovative 'body logic' of its IRON humanoid robot. This deep-dive, released recently, highlights significant advancements in the robot's design, including a revamped reinforcement learning framework and proprietary algorithms that enable the simulation of intricate 3D-printed lattice muscles. The development aims to enhance the robot's functionality and adaptability, showcasing XPENG's commitment to pushing the boundaries of robotics technology. This initiative reflects the company's broader strategy to integrate advanced machine learning techniques with cutting-edge manufacturing processes, positioning the IRON humanoid as a leader in the evolving robotics landscape.
China XPeng IRON
AI Is Designing Radio Chips That Humans Couldn’t Even Imagine
Researchers at Princeton University have made significant strides in the design of radio-frequency integrated circuits (RFICs), a critical component for advancing wireless technologies such as 5G, autonomous vehicles, and satellite communications. Utilizing reinforcement learning and inverse design techniques, the team has developed a method to create RFICs from scratch, drastically reducing design time and achieving record performance levels. This innovative approach leverages AI to navigate the complex design space of RFICs, traditionally seen as an art requiring years of expertise. By employing machine learning algorithms, the researchers can generate novel circuit layouts that outperform existing designs while minimizing the time taken for development. The project, which began after the success of AI in games like Go, aims to overcome the limitations of conventional RFIC design, which has remained largely artisanal. The researchers emphasize the need for large, shared datasets and open ecosystems to further enhance AI's capabilities in understanding electromagnetic and circuit behaviors. As the demand for advanced RFICs grows, the potential for AI-driven design to revolutionize the field is becoming increasingly apparent. The findings have attracted attention within the RF community, sparking discussions about the future of AI in circuit design and the importance of collaboration between AI researchers and chip designers to unlock new possibilities in technology.
Machine-learning Ic-design Chip-design Rf Rfic
Amazon's ResMimic Teaches Humanoids to Handle Objects by Adding Precision to General Motion
Amazon has unveiled its latest robotics initiative, ResMimic, which aims to enhance the capabilities of humanoid robots in performing complex loco-manipulation tasks. This innovative project employs a two-stage residual learning framework that allows for the efficient teaching of these skills. By refining a general motion policy with specific corrections tailored to individual tasks, the system empowers a Unitree G1 robot to adeptly manage heavy and irregular objects with remarkable precision. This development is part of Amazon's ongoing efforts to advance robotics technology and improve automation processes in various applications.
reinforcement-learning Unitree Robotics AI Amazon robotics
KAIST Unveils Advanced Four-Legged Robot with Autonomous Navigation Technology
KAIST's mechanical engineering team, led by Professor Park Hai-won, announced a breakthrough in robotic technology on July 16. They developed a four-legged robot capable of autonomously selecting and switching between various gaits in real-time, enabling it to navigate complex outdoor environments with speed and stability. This innovation is significant as it integrates a new control architecture called APT-RL (Action Pre-training Reinforcement Learning based on Transformers), which allows the robot to learn movement through computer simulations rather than traditional motion capture. The robot, named KAIST HOUND, demonstrated its capabilities by traversing diverse terrains, achieving peak speeds of 6 meters per second, faster than an average cyclist. Future developments to watch include the potential applications of this technology in disaster response, defense tasks, and industrial inspections. The research was published in the July issue of the journal Science Robotics, highlighting its importance in advancing the field of robotic control and physical AI.
Four-Legged Robots Robotics Technology AI Autonomous Navigation
Korean Researchers Develop AI Framework for Robot Dog's Adaptive Movement in Complex Terrain
Researchers from Korea have created an AI framework that allows a quadruped robot to autonomously adapt its motor skills while navigating challenging environments. This system enables real-time gait adjustments for traversing forests, climbing stairs, and overcoming obstacles using only onboard sensors and computing capabilities. The significance of this development lies in its potential applications for autonomous search-and-rescue and exploration missions. The Action Pretrained Transformer-based Reinforcement Learning (APT-RL) framework enhances agility by combining pretrained locomotion skills with adaptive decision-making, demonstrating the robot's ability to handle diverse obstacles effectively. Future observations will focus on the framework's deployment in real-world scenarios, as it has already shown impressive performance on KAIST’s quadruped robot, HOUND. The robot's ability to switch between different gaits based on terrain and speed, achieving speeds of up to 6 meters per second, highlights the effectiveness of the APT-RL approach in complex environments. No further timeline was disclosed at the time of publication.
AI and Robotics
KISS Institute Launches BotBall to Enhance STEM Education with Student-Led Robotics
The KISS Institute for Practical Robotics (KIPR) has introduced BotBall, a robotics program designed to foster creativity and critical thinking among students. This initiative emphasizes student-led engineering, allowing participants from elementary to high school to engage in hands-on learning using a standardized kit. The program ensures a level playing field by providing all teams with the same materials, promoting accountability and project management skills without adult intervention during competitions. BotBall challenges traditional educational models by integrating real programming languages like C and Python into its curriculum, demonstrating that students can handle complex coding at an early age. The Junior Botball Challenge (JBC) further innovates by allowing up to five students to collaborate on a single robot, shifting the focus from competition to inquiry-driven problem solving. This approach encourages teamwork and a deeper understanding of both mechanics and software among participants. As the school year approaches, KIPR is expected to release more details about the upcoming competition schedule. The BotBall program represents a significant shift in STEM education, moving away from conventional roles and fostering a new generation of students who are well-versed in both engineering and programming disciplines. No further timeline was disclosed at the time of publication.
Autonomous Mobile Robots (AMRs) Educational News competition education workforce
SoftServe Introduces Virtual Gyms for Enhanced Robotics Training and Deployment
SoftServe has highlighted the importance of 'virtual gyms' for robotics teams, emphasizing their role in preparing robots for dynamic environments. These high-fidelity simulation environments allow robots to train, fail, and recover safely before real-world deployment, addressing the challenges posed by unpredictable operational conditions. The global robotics market is projected to grow at a 19.6% CAGR from 2026 to 2036, underscoring the need for effective training solutions like virtual gyms to enhance robotic autonomy and performance. The shift from programmed automation to physical AI necessitates that robots adapt to constantly changing environments, which traditional training methods struggle to accommodate. Virtual gyms integrate technologies such as digital twins, reinforcement learning, and sensor modeling to provide a comprehensive training platform. This approach mitigates the risks and costs associated with real-world trials, enabling teams to generate valuable training data in a controlled setting, thus improving deployment success rates. Looking ahead, the adoption of virtual gyms is expected to become a standard practice in robotics development, as they offer a solution to the simulation-to-reality gap. No further timeline was disclosed at the time of publication, but the increasing complexity of robotic tasks suggests that the demand for such training environments will continue to rise as the industry evolves.
Artificial Intelligence Artificial Intelligence / Cognition Autonomous Mobile Robots (AMRs) Development Tools / SDKs / Libraries Industrial Robots Logistics
Exploring Automation Intelligence: Opportunities and Challenges in AI for Manufacturing
The manufacturing sector is experiencing a surge in artificial intelligence (AI) applications, driven by recent advancements in speech, language, and content generation technologies. Engineers and technology leaders are keenly observing these developments to enhance quality, minimize rework, and increase throughput. However, many organizations face challenges in translating AI demonstrations into tangible business value, revealing the complexities of deploying AI in production environments. Despite significant investments in AI and machine learning (ML), the manufacturing industry is encountering hurdles similar to those faced during the initial wave of data science and ML in the context of Industry 4.0. Many early projects failed to deliver operational value due to the misalignment of algorithms designed for consumer behavior with the deterministic needs of industrial settings. As manufacturers increasingly seek actionable insights from their data, the need for a deeper understanding of AI technology and its application in industrial contexts becomes critical. Looking ahead, the emergence of automation intelligence, which integrates lessons from past experiences with current AI tools, offers a promising framework for addressing complex industrial challenges. As AI technologies like generative AI and foundation models continue to evolve, their successful implementation will depend on ensuring real-time grounding, safety, and regulatory compliance in manufacturing processes. No further timeline was disclosed at the time of publication.
Factory / Workforce
AI Models Overthink Problems—and It’s a Security Risk
Large language models (LLMs) that can think through problems step-by-step have significantly increased the scope of tasks that AI can tackle. But new research suggests these reasoning capabilities also introduce a critical vulnerability that could allow attackers to slow these systems to a crawl.While earlier generations of LLMs would immediately produce a response to a user’s request, today’s most advanced models generate an internal monologue where they break down the problem into steps and reason about the best way to tackle it before providing an answer. This has allowed AI to tackle increasingly complex problems, particularly in areas like coding and math.However, previous research has shown that these models are susceptible to sometimes producing excessively long streams of reasoning that do little to boost performance, a phenomenon known as “overthinking.” In research presented this week at the International Conference on Machine Learning 2026 in Seoul, researchers from Zhejiang University and e-commerce giant Alibaba in China demonstrate that they can deliberately induce overthinking by subjecting models to logically inconsistent prompts. The result is a form of denial-of-service attack on commercial AI models.Evolutionary Prompt Attack on LLMsThe team has developed an evolutionary algorithm that corrupts the logical structure of prompts, causing models to spiral into overthinking as they attempt to reason through fundamentally unsolvable problems. Generating longer responses costs more and increases the load on a model provider’s servers, so if done at scale, the researchers say, this could significantly degrade the experience of legitimate users. The attack was effective against reasoning models from leading AI companies including DeepSeek-R1, Alibaba’s Qwen3-Thinking, OpenAI’s GPT-o3, and Google’s Gemini 2.5 Flash and resulted in outputs up to 26 times as long as standard responses on a standard math benchmark.“Across multiple datasets and reasoning models, our method substantially amplifies the output length,” Wei Cao, a masters student at Zhejiang University, wrote in an email to IEEE Spectrum. “Our results suggest that overthinking is not an isolated phenomenon specific to individual models, but rather a shared vulnerability among modern reasoning models.”The team’s approach builds on previous research from another group of researchers that showed reasoning models tend to overthink when faced with a question in which a key premise has been removed—such as asking how far someone who walks ten miles a day covers in total without specifying how many days they walked for. Rather than identifying that the problem is unsolvable, models often engage in extended but ultimately fruitless reasoning loops in an attempt to answer the question.Taking the idea a step further, the authors took 940 problems from three math benchmark datasets and used an LLM to break down their logical structure into a set of premises and a final question. The genetic algorithm then jumbled these up using a variety of “mutations,” including swapping premises between problems, adding extra premises to problems, deleting existing premises from problems, and swapping the final questions between two sets of premises.After each round of mutations, the problems are scored on how many words they cause a target model to output and also whether they increase the frequency of specific linguistic markers of overthinking—words like “but,” “wait,” “maybe,” or “alternatively.” The problems that scored highest on both measures are retained and the remaining ones are jumbled up again, and this process is repeated for five generations. Crucially, the approach doesn’t require access to the internals of a model and can generate malicious prompts by simply querying the target, which makes it possible to attack closed-source commercial services, says Cao.Overthinking Vulnerability in AI ModelsThe researchers found that the approach consistently led to outputs several times longer than those generated by the unmodified questions for the reasoning models they tested it on. The biggest jump came from DeepSeek-R1 on the MATH dataset, which is made up of problems from high school math competitions, where the maximum output was 26.1 times as long as the longest response the model provided to unaltered questions. While the main thrust of the research was focused on math problems, the authors also tested it on coding, scientific reasoning, and dialogue challenges, and observed significant jumps in output length in all three.One challenge for the approach is that developing the malicious prompts requires repeated queries to expensive reasoning models, which Cao admitted could limit its cost-effectiveness. However, the researchers also demonstrated that when they used a smaller, cheaper model to generate the malicious prompts they were still able to induce the target models to produce outputs several times longer than normal. This ability to transfer malicious prompts between models significantly increases the attack’s feasibility, Cao wrote.However, he pointed out that the goal of the research is not to develop a practical DoS attack on reasoning models. Factors like the providers’ pricing model, rate limiting policies, context window size, and existing defenses could all impact how effective the approach is. The intention is instead to highlight these models’ vulnerability to logically inconsistent prompts so that providers can attempt to mitigate the problem.“Our objective is not to demonstrate that large-scale attacks can be launched at negligible cost, but rather to establish that this attack surface exists,” he wrote. “Our results indicate that the vulnerability represents a realistic security concern.”
Llms Artificial-intelligence Denial-of-service Cybersecurity
An AI-powered control system for robots with legs
Walking robots, such as quadruped robotic dogs, must be able to move safely through rough, often changing environments. Today, there are two main ways to program these walking, or legged, robots. The first is called model predictive control. This technique optimizes the robot's behavior but relies on accurate dynamics models, which are challenging to achieve in real-world settings and often require simplifying assumptions. The second is model-free reinforcement learning, which allows the robot to learn reliable but fixed behaviors, making them difficult to adapt after training.
Robotics
Small-AI Models Gain Traction Around the World
One morning in 2019, Adebayo Alonge was in a Cape Town hotel room, preparing to demonstrate his startup’s AI answer to a serious problem in African health care: counterfeit medication, which kills thousands of people across the continent every year.The RxScanner is a handheld spectrometer that scans a pill with infrared light, then sends the item’s molecular profile to an AI model equipped with a pharmaceutical database. In seconds, the AI identifies the medication from its molecular profile—or reports that it’s phony.Pharmacies were using the system in more than a dozen countries, including Ghana, Kenya, Myanmar, and Alonge’s native Nigeria. But that morning in South Africa, it didn’t work. “I was shocked,” Alonge says.The spectrometer connected to the AI model—but the data center was 14,000 kilometers away and bandwidth was limited. “Our server was in the United States, and just to get the result of a single scan was taking me over 5 minutes.”So Alonge immediately asked his engineers to shrink the AI model down to a smaller, low-power, unconnected version that could run entirely on his Android phone. They produced it 2 hours later, and that saved the demo.More importantly, the work birthed a new version of his device, which can authenticate a pill in places without broadband, computers, or even reliable electricity. It also turned Alonge into an advocate for this kind of “small AI.”Small AI for Global Health Care AccessSmall AI is a far cry from wealthy nations’ colossal large language models (LLMs), hyperscale data centers, multibillion-dollar investments, and debates about AI consciousness. But for millions of people around the world, the only AI that matters, and often the only kind available, is small. (According to a World Bank Report issued in November, only 0.7 percent of internet users in the world’s poorest countries have used ChatGPT, compared to a quarter of all internet users in the most developed nations.)“Most people are discussing AI from the LLM/generative side. But that needs a lot of computing power, electricity, massive data, and skilled people to manage it,” Ajay Banga, president of the World Bank, said last January at the World Economic Forum, in Davos. “Outside the developed world, other than maybe India and China, very few countries have that combination.”By contrast, small AI can deliver useful, even life-saving services to people in areas that have none of those things, Banga said. In India, where the government’s AI plans call for more development of small AI, many such systems are working for farmers.For example, a drone-based system developed by Bala Murugan and colleagues at the Vellore Institute of Technology, in India, takes photos of cashew plants and quickly identifies those with splotches that indicate disease. All the processing takes place on the drone itself, so there’s no need for a computer on-site, nor for a connection to a central server.Using small language models trained for a specific problem, and sometimes running on cheap, low-power devices, other small-AI implementations have been developed to identify ant infestations in a Uruguayan vineyard, detect the presence of malaria-carrying mosquitoes in a number of nations, and run electrocardiograms from an Arduino device in parts of Brazil that lack access to more complex equipment.“This is the most important area in AI nowadays,” says Marcelo José Rovai, a professor at the Institute of Engineering and Information Systems at the Federal University of Itajubá, in Brazil, who was involved in all three projects. “It’s growing very fast.”Low-Power, Small-AI Models on Devices Small AI models can run on a variety of low-power devices, including [from left to right] an Arduino Nano 33 BLE Sense, a Seeed Wio Terminal, and an Arduino Portenta.Moez AltayebFor Alonge, Rovai, and other advocates, small AI is not just “a promising trend,” as that November World Bank report calls it. It may be, in the long term, the form of AI that will touch the most lives and remain sustainable after some of the giant models become too costly for most users.“I think the future of AI is not like one giant model, at a center. I think it’s millions of small, precise models deployed at the edge, each one solving like a specific problem, a specific context,” Alonge says. This is partly because much of humanity—including people in parts of rich countries as well as the developing world—lives without access to cutting-edge frontier models. But, he says, it’s also because those models are not sustainable.“If someone is not subsidizing it, most people will not be able to afford those models. So those of us who are said to be small-AI developers are the ones who will have to build for the majority of the world,” Alonge says.There is no strict definition of “small AI,” but people often use the term for language models with at most a few billion parameters. (Compare that to cutting-edge models, which can include more than a trillion.) That’s small enough to run directly on a phone or a Raspberry Pi. That’s what allows these applications to run on devices without a connection to a data center and use only a few watts of power, often supplied by a battery or a solar panel.Despite their small footprint, these models aren’t fundamentally different technology from that of gigantic AI models, Rovai says. Many instances of small language models were created the same way the phone-based version of Alonge’s pharmaceuticals scanner was—by “pruning” large models, or removing the parameters that weren’t involved in the task. The result is a system that’s less capable generally but still very good at the specific job it was pruned for, Rovai says. A lighter version of RxAll’s RxScanner spectrometer sends its results to an AI model run locally on a phone to check that a drug’s molecular signature is genuine.RxAllOther small models are created by “distillation.” They are trained to mimic a large model, until their performance approaches that of their “teacher,” Rovai says. In other cases, a larger model’s precision is reduced, for example, so that a model run on 32-bit architecture can run on 8-bit designs. In situations where the machine learning application is being used to classify data or predict patterns (like an ant infestation), it’s trained from the beginning on a small device, not derived from a larger model at all. Running all these small, specialized systems is becoming easier, Rovai says, for two reasons.The first reason is that hardware is getting better and more capable while using less power, he says. This means more and more phones can run small AI—especially those equipped with neural processing units, which are specialized chips that handle AI tasks like facial recognition and changing the brightness, shadows, or contrast in a photo.In 2025, slightly more than a third of all smartphones shipped worldwide were capable of running generative AI, and that figure will reach 45 percent by the end of this year, according to the technology research firm Counterpoint. By the end of next year, slightly more than half of all smartphones will be able to run a small AI model.The second reason Rovai cites is the shrinking footprint of language models. Both Google DeepMind’s Gemma 4 (released in April) and Alibaba’s Qwen 3.5 are “fantastic” for small AI, Rovai says. Both models are “open weight,” meaning users can adjust the connections between parameters to suit their needs. This makes it easy, for example, “to take a lot of data from, say, the milk industry and retrain the model specifically on that,” Rovai says.Rovai illustrated these reasons on a Zoom call, using one of his most recent experiments. Holding up a device, he says, “This is the new Arduino UNO Q—a US $50 device with a Qualcomm chipset. I’m running a language model here, which collects data from sensors and analyzes that data to detect tiny pools of water where mosquitoes might be breeding. It takes 3 watts to run it.”Support for Small-AI DevelopmentConvinced that millions of people are already benefiting from these kinds of applications, the World Bank now actively promotes small AI with grants, mentorship programs, financing, technical advice, and models of government policies that are friendly for small-AI development. For example, in Rwanda, the World Bank is backing a government program to help low-income households get devices that can run AI.All that said, no one claims that large language models are going away entirely. To create a generative AI that can run on a phone or other small device requires the architectural insights, data processing, and results of a larger model, Rovai says. “We need the big models to create these smaller models.” And for all that small AI can benefit people without access to big AI, the technology can’t solve the larger problems of development and digital inequality, Alonge says. Implementing small AI won’t allow nations to escape the challenge of creating an ecosystem to support AI: reliable power, a supply chain that works, and an educational system that develops the talents needed to create AI tools.Though his drug-scanning system can run for days on a phone with no connection, “you still want to be able to enable periodic syncing for updates with new signatures for the medications and analytics,” Alonge says. “And even when you are using batteries, reliable power is important. That phone battery is not going to last forever.”In many parts of the world, the future of small AI isn’t assured, he says. “It works, and many places will eventually need to use it. The question is whether or not the political actors are wise enough to invest in infrastructure to support it long term.”
Small-language-models Artificial-intelligence Llms
Tsinghua alumni launch AI startup, securing hundreds of millions in angel funding for the automotive industry.
Embodied intelligence company Guangxiang Technology has successfully secured hundreds of millions in angel funding, with significant participation from leading investors including Zhuhai Technology Industry Group, Xingsheng Capital, and several others. Founded in April 2025 through a collaboration between Tsinghua University's Vehicle and Transportation Institute and its AI Institute, Guangxiang aims to advance the development of its physical native base model and commercialize embodied intelligent robots for industrial applications. The company's founder and CEO, Zhang Tao, previously led the spatial perception engine at Amap, while co-founder Professor Li Shengbo is a renowned expert in reinforcement learning and autonomous driving. The team, which includes members from major tech firms like Alibaba and Huawei, is focused on a unique technological approach that diverges from mainstream visual-language-action models. Guangxiang's flagship product, the Phi-Bot X1, was launched in June 2026 and is designed for industrial environments. The robot has demonstrated impressive capabilities, completing a full welding operation on an automotive production line without errors during a 21.5-hour continuous run at the 2026 ATC exhibition. The company plans to expand its applications in the automotive sector, targeting the remaining 30% of automation gaps that traditional methods cannot address. Zhang envisions a robust market for automotive production line robots, estimating a potential market size of around 100 billion yuan in China. Guangxiang Technology is committed to refining its products and scaling operations, with a focus on real-world applications and continuous feedback to enhance its offerings.
Robotics needs a service framework.
RSF defines a common language for robot service capability, lifecycle operations, certification pathways, and service-provider networks.
