Manufacturing Technology Centre (MTC) has launched a new Robot Experience Centre (REC). This state-of-the-art, vendor-neutral facility is situated at MTC's Ansty Park location in Coventry. Its primary goal is to accelerate the adoption of robotics and automation technologies within UK manufacturing. The REC allows manufacturers to practically explore, test, and validate various robotic solutions. These real-world testing capabilities support informed investment decisions for industry adopting advanced automation.
Casey Harrell, who has ALS and is paralyzed, uses electrodes implanted in his brain for communication. He was fitted with the device following an operation in July 2023 after being diagnosed with ALS. The system records activity from the speech motor cortex, translating neural signals into discernible phonemes and words. Since initial use in 2023, Harrell has gained substantial independence, utilizing the BCI for communication, web browsing, and working. He has reportedly used the device over 3,800 hours at home without researchers present, establishing him as a "power user" of speech BCIs.
Pegasus Tech Ventures has launched a \$60 million investment fund specifically targeting physical AI startups. Investment areas will include robotics, healthcare technology, automation, and intelligent systems. The venture's funding aims to support companies building real-world applications of artificial intelligence. Pegasus and CYBERDYNE are partners backing this new initiative. Details regarding the launch were initially published on The Robot Report.
Kawasaki Robotics plans to showcase its approach to intelligent automation at the Automate event. Highlights of their demonstration include an 8-DoF robot and Pulseboard inspection technology. Central to their unveiling is the debut of the RL030N physical AI platform. This presentation aims to illustrate their vision for advanced robotic systems in industry. The initial reporting on this announcement was published by The Robot Report.
The "MIGHTY" system is an open-source technology capable of developing optimized path plans for robotic movement. It functions by rapidly generating paths that ensure both efficiency and obstacle avoidance. This capability has significant potential applications in various sectors. Specifically, it can help streamline complex operations such as disaster recovery efforts. Furthermore, the system promises to improve logistics through more efficient parcel delivery routes.
Genesis AI has unveiled Eno, marking their debut general-purpose robot model. Unlike many humanoid designs, Eno features a wheeled base instead of two legs. It is equipped with human-scale dexterous hands and boasts an appliance-like aesthetic. This unveiling represents a conscious challenge to the traditional bipedal format in robotics. Additionally, Genesis AI announced a new strategic deployment partnership with LG CNS.
Researchers have developed a novel spatial memory system designed for use in robots. This technology allows robots to efficiently capture detailed information about objects within an environment during exploration. By building a detailed map and object registry, the system can theoretically pinpoint lost items. Such advanced memory capabilities could allow robots to assist humans by locating misplaced belongings. This represents a significant step toward more capable robotic assistants.
Developing a ROS 2 robot requires connecting an STM32 microcontroller running Zenoh-native Zephyr RTOS to an SBC utilizing rmw_zenoh on a Zenoh router. The author is evaluating three communication options: micro-ROS, Zenoh-Pico with rmw_zenoh_pico, and Pico-ROS. Concerns exist regarding the suitability of micro-ROS given existing Zenoh usage, the portability of Zenoh-Pico to STM32/Zephyr, and the lack of documentation for Pico-ROS on this hardware combination. Ultimately, the author seeks practical experience or alternative architectural suggestions for integrating these systems.
Raymond Corporation has partnered with Third Wave Automation to expand physical AI capabilities within its automated lift truck fleets. This technology collaboration aims to bring advanced, AI-enabled automation across Raymond's fleet. The partnership leverages several years of joint development and operational experience that began in 2021. Development efforts have previously received backing from Toyota Ventures. Ultimately, the alliance seeks to enhance the functionality of Raymond's lift trucks using cutting-edge AI technology.
RLWRLD was designated a World Economic Forum Technology Pioneer for 2026. The company specializes in physical AI and developed the proprietary Robotics Foundation Model named RLDX-1. This selection places RLWRLD among 100 innovative firms each year recognized by the WEF. These pioneering companies are expected to cause long-term, paradigm-shifting impacts across global industries and society.
Kodiak employs AI and probabilistic risk assessment to quantify the safety of autonomous trucking technology. The development pace of such technologies is accelerating rapidly due to advances in artificial intelligence. Achieving high standards of safety requires rigorous measurement methods rather than solely relying on large budgets or extensive fleet sizes. Instead, the focus is on extracting the most precise insights for demonstrable safety proof. This approach demonstrates a commitment to deep, measurable safety validation in autonomous transport.
X Square Robot is focusing on enabling robots to function effectively within unpredictable, real-world environments rather than solely emphasizing flashy feats like backflips or dancing. Their goal aims to build the crucial "brain" required for embodied artificial intelligence. This technology seeks to move robotics beyond controlled demonstrations into practical daily applications. The company's focus suggests a shift toward solving complex operational challenges in messy human settings.
ANYbotics, a provider of AI-driven robotic inspection solutions, has established a new engineering and AI hub in Barcelona, Spain. This facility is situated at the DFactory Barcelona campus. It marks the company's third global location, complementing existing sites in Zurich and San Francisco. The expansion aims to support increasing demand for ANYbotics' autonomous inspection solutions. These solutions are targeted toward asset-intensive and process industries globally.
Industrial automation growth is necessitating rapid changes in infrastructure within factories and shipping centers. These evolving environments require tools that enhance production, minimize errors, and reduce waste for efficient goods movement. However, the advanced technologies driving this—including robots, AI, and autonomous machines—demand sophisticated supporting systems. The challenge lies not just in implementing the hardware but in powering the necessary intelligence. Therefore, upgrading the underlying "brains" supporting these complex automated systems is crucial for sustained growth.
Automatic generation of populated robot ontologies from Unified Robot Description Format (URDF) models is presented to address the bottleneck of manual ontology construction. Because URDF files lack inherent semantics beyond structure, Large Language Models (LLMs) are utilized to infer meaningful relationships by prompting them with established ontological concepts. The proposed pipeline enhances reliability through majority voting across multiple LLM queries and incorporates syntactic and schema-level validation. This method effectively converts low-level kinematic descriptions into structured, semantically rich knowledge representations. Initial evaluations show the approach successfully bridges the gap needed for explainable reasoning in human-robot interaction.
Researchers introduce HRDX, a new, large-scale vector HD-map dataset spanning approximately 40 hours of driving across 1,400 km. The data is captured using six synchronized cameras, a 128-beam LiDAR, centimeter-level RTK GNSS/IMU, and aligned aerial orthoimagery. HRDX features annotations for 10 vector map classes and over 20 semantic and topological attributes. Evaluation demonstrates that the dataset's large scale improves online vector-map construction. Furthermore, incorporating aerial imagery provides a structural prior that enhances geometric map quality across various training setups.
ParkingTransformer is a novel framework designed for end-to-end autonomous parking by integrating Large Language Models (LLMs) with multi-view perception. It addresses limitations in existing methods by using LLM features and trajectory queries directly from sensor data, bypassing dense Bird's-View representations. To enhance spatial reasoning, the method incorporates 3D positional encoding, and it uses a fixed-window streaming mechanism for efficient historical processing. A coarse-to-fine decoding strategy is employed to refine the generated planning trajectories progressively. Experiments on CARLA and real platforms showed promising results, achieving an 88.70% average success rate in the field.
ACE-EGO-0 is a novel framework designed to unify Vision-Language-Action (VLA) pretraining by integrating heterogeneous human and robotic data. It addresses the difficulty of joint training due to disparities in action spaces and embodiment structures. A key component converts large amounts of raw egocentric human videos into robot-compatible pseudo-action trajectories via a scalable pipeline. This framework utilizes a unified camera-space action representation and employs a reliability-aware objective, incorporating a human auxiliary loss for noisy supervision. Evaluations show that using this combined data significantly boosts performance on benchmarks like RoboCasa GR1 TableTop, achieving state-of-the-art results.
Existing vision encoders for robotics suffer from data scarcity, lacking the paired vision and action data needed for visuomotor control. To address this, authors propose CAIP (Contrastive Action-Image Pre-training), which uses human hand poses from large egocentric videos as a proxy for robot end-effector actions. This method extracts 3D hand keypoints and learns a unified action-image representation using a contrastive objective. Training on extensive human video alongside limited robotic data, CAIP outperforms state-of-the-art encoders like DINOv2 and SigLIP. The technique achieves over 30% performance gains on difficult real-world dexterous manipulation tasks, demonstrating scalability for physical interaction representations.
VISTA improves Vision Navigation Foundation Models by conditioning predictions on normalized action histories to mitigate physical inaccuracies arising from action normalization. Integrating a DINOv3 encoder enhances representation learning, allowing the model to capture spatial and geometric details even in visually repetitive settings. This approach addresses the critical vulnerability of current models when translating normalized actions to real-world robotics. The resulting framework generalizes robustly to out-of-distribution environments. Testing demonstrated 100% goal prediction accuracy and an average of 95% checkpoints crossed across unseen outdoor, forest, and office settings.
Real-time trajectory generation for robotic servicing near tumbling targets is difficult due to coupled nonlinear dynamics and safety constraints. A proposed framework addresses this by decomposing the problem into translational and attitude/torque allocation stages. It applies a causal transformer warm-start specifically to the computationally intensive, coupled attitude-manipulator torque stage. Testing showed that this learned warm-start reduced sequential convex programming (SCP) iterations by up to 28% and runtime by 23% for cost-optimal approaches. Furthermore, when used for nonconvex feasibility projection, it nearly halved the runtime while maintaining superior trajectory robustness compared to heuristic initialization.
Contactless respiratory rate (RR) monitoring is vital for safe remote triage in emergencies but faces challenges from varied illumination and sensor limitations. This paper introduces a modality-adaptive framework utilizing multiple sensors—RGB, thermal, NIR, and low-light cameras—on edge-computing mobile robots. The system incorporates brightness adaptation, keypoint guidance for robust region extraction, and an SQI filter for accurate estimation. Evaluation across diverse robotic platforms confirmed the framework's generalizability without platform-specific tuning. Results defined operational ranges, noting RGB offers the broadest coverage, while specific modalities are suited for different distance and lighting conditions.