OpenAI released the Codex Micro, a tangible $230 physical keypad. This device is presented as part of an update to OpenAI's offerings. The report indicates that this product launch signals a move in AI hardware integration. It suggests physical interaction might be becoming a necessary component for advanced AI applications.
Heat pumps are becoming increasingly popular due to their efficiency in providing heat using electricity. These systems offer a modern alternative to traditional heating methods. Despite the uncomfortably hot summer weather, their growing adoption suggests a shift in residential HVAC technology. Their use of electricity makes them an efficient heating option as climate concerns grow.
Agility, through its CEO Peggy Johnson, advocates for a comprehensive policy framework within the United States for the humanoid robotics sector. They outlined six specific recommendations to bolster the current operational environment for the industry. These suggestions aim to strengthen existing positive developments surrounding humanoid robots. Implementing such policies is considered necessary for the continued growth and safe deployment of this technology.
NVIDIA launched the T3000 and T2000 modules to address growing demand for powerful, compact AI computing in real-world applications. These new products are built upon the advanced NVIDIA Thor architecture. They are designed to support general-purpose robots and autonomous machines moving out of research settings. Specifically, they provide the necessary power efficiency for running large foundation models directly at the edge. This aims to accelerate mass-market adoption of robotics and edge AI technologies.
The Netherlands is pioneering advanced technological integration into its armed forces using uncrewed systems. The Royal Netherlands Navy is deploying autonomous vessels, such as "Defender 1" and "Defender 2," which function remotely via computer control. These unmanned platforms are designed to operate in dangerous zones, keeping personnel safe from risk. Experts suggest these technologies represent the future direction for naval operations. Current testing missions near Den Helder aim to fully evaluate the capabilities of these human-remote systems.
Walden Robotics has launched after developing and emerging from stealth mode. The company focuses on creating full-stack physical Artificial Intelligence. It announced its launch with a valuation of $1.1 billion. This financial information was reported by The Robot Report. Walden's primary focus is on general-purpose robots.
Icarus Robotics is utilizing KULR technology for a specific application. KULR will supply its KULR ONE Space battery systems. These batteries are designated to power the onboard systems of JOY. JOY is described as a free-flying space robot. This information was originally published on The Robot Report.
Environmental monitoring using UAVs necessitates route planning that maximizes coverage while respecting energy and operational limits. This paper presents preliminary results from a systematic literature review following the PRISMA 2020 framework, searching databases for studies from 2015 to 2026. The review focuses on path, coverage, and informative planning, analyzing algorithmic families and associated metrics. Initial findings show significant research emphasis on coverage-oriented approaches and multi-UAV coordination. However, the literature also indicates a scarcity of work addressing weather uncertainty or complex obstacles, suggesting a need for structured synthesis to advance realistic field applications.
Existing vision-language-action (VLA) benchmarks fail to model complex human-robot collaboration structures like intent understanding and temporal synchronization. HRIBench introduces a diagnostic benchmark using structured scenario scripts that explicitly model agent roles and coordination constraints for collaborative tasks. The benchmark defines three interaction roles—Instructor, Collaborator, and Intruder—across 13 task types with over 650 evaluation episodes. It employs interpretable metrics measuring synchronization, responsiveness, protocol compliance, and safety beyond simple success rates. Evaluation using HRIBench reveals that current foundation robot policies possess limitations in temporal coordination and intent-aware behavior within collaborative settings.
High-Voltage Electrostatic Actuators (HVEAs) are presented as a promising solution for haptic interfaces due to their soft, thin, silent, and energy-efficient nature compared to conventional actuators. The paper reviews four main classes of HVEAs: electrostatic switchable adhesives, dielectric elastomer actuators, soft electrohydraulic actuators, and electrokinetic pumps. For each type, it details the mechanism, bandwidth, force density, and scalability for providing haptic feedback. This cross-technology analysis identifies shared design limitations and strategies for enhancing ergonomics and fabrication. Ultimately, it outlines HVEAs' unique potential to advance practical, wearable haptic interaction systems.
Embodied AI heavily relies on massive parallel simulation using GPU-accelerated tools, but key trade-offs like efficiency, fidelity, and determinism are understudied at scale. The research introduces GPUSimBench to expose hidden limitations in current robotic simulators (e.g., Isaac Lab). This benchmark evaluates physical consistency using an inclined-plane task to compare simulation dynamics with reality. Furthermore, it measures parallel scalability by assessing throughput across increasing environments. Crucially, the work quantifies inherent non-determinism in GPU batch execution, showing variability even when conditions are identical.
Proposes a hierarchical Bayesian generative model to operationalize existing, non-algorithmic guidelines for designing humanoid robots within the context of the uncanny valley. The model represents human affinity as posterior-weighted negative category-conditional surprise, linking ambiguity and mismatch to increased surprise. It maps uncanny-valley mechanisms onto four measurable variables: deviation from predicted mean, cross-modal inconsistency, prediction uncertainty, and observational uncertainty. Simulations demonstrated that ambiguity and appearance-motion mismatches reduce affinity, while uncertainty significantly reshapes the valley's effects. Human-subject experiments confirmed this, showing that greater observational uncertainty mitigated familiarity decreases at intermediate likeness levels.
A novel teleoperation architecture addresses limitations in subsea ROV operation using a ROS-Unity framework to separate planning from avoidance. It employs a Dynamic Adaptive Viewpoint System (DAVS), utilizing 3D Gaussian Splatting, to provide occlusion-free exocentric views rather than standard 2D feeds. Sensory load is further managed by mapping obstacle clearance to intuitive haptic cues via a vibrotactile suit. Testing on 30 subjects revealed that the 3DGS-driven exocentric view excelled under high communication latency (0.5–1.0 s). Additionally, fNIRS data showed that proactive spatial context from DAVS mitigated cognitive overload associated with increased latency in traditional egocentric control.
To overcome the prohibitive cost of manual data collection for open-world mobile manipulation, a synthetic data engine called WANDA is introduced. WANDA reconstructs background Gaussian splats and robot-object interaction trajectories from single RGBD observations to create a foundational world substrate. It then reconfigures contact-rich segments using whole-body motion planning to generate vast numbers of diverse spatial configurations. To boost robustness and generalization, the method employs Corrective State Expansion and synthesizes data across varied 3D worlds derived from everyday photos. Experiments demonstrate that policies trained with WANDA achieve strong long-horizon, spatial, and cross-environment generalization, even supporting zero-shot deployment on different robot morphologies.
Accurate attitude estimation is challenging for autonomous vehicles, especially without GNSS or during high acceleration when IMUs alone fail. This paper presents two barometer-aided architectures to enhance estimation by utilizing altitude data for vertical motion information. The first approach designs a deterministic Riccati observer cascaded with a complementary filter, ensuring almost-global asymptotic stability under uniform observability. Secondly, a novel nonlinear observer is proposed that integrates IMU inputs with barometer and magnetometer outputs within an SO(3)xR2 framework, guaranteeing local exponential stability. Validation using simulated and real data shows this method offers a practical and reliable sensing modality when conventional velocity measurements are unavailable.
A minimum-lap-time planning framework incorporating robustness is presented for motorsport applications. This methodology extends prior work by managing both state disturbances and uncertainty in vehicle parameters like moment of inertia and drag coefficient. To maintain computational feasibility, a spatially selective, parsimonious strategy limits robust constraints to critical circuit segments. Validation was performed using an MPC as a virtual driver on a simulated FSAE car over 1000 runs with random disturbances. The resulting robust references demonstrated improved driveability, exhibiting fewer failures and better signal dispersion compared to nominal plans at a minor time cost.