Researchers developed a larval zebrafish-inspired robotic platform (ZBot) to investigate the energy efficiency of intermittent swimming compared to continuous swimming. They found that intermittent bout-and-glide swimming reduces energy consumption across all achievable velocities, especially in viscous fluid regimes. The team proposed the "actuator efficiency" hypothesis, suggesting that intermittent swimming enhances the transfer of energy from electricity to kinematic tail movements, allowing actuators to operate in more energy-efficient regimes.
Persona AI has developed a humanoid robot to focus on welding in shipyards, targeting long, linear welds that are relatively uncomplicated. This task leverages the robot's ability to perform millimeter-scale repeated motions, which robots excel at, particularly over long periods of time. Shipyards offer a compelling environment for the humanoid robot, as it can work on the ground, overhead, and through portholes, and the robot's safety concerns are lessened due to the industrial environment. The company aims to create high-fidelity simulations and collect demonstration data to expand its applications to grinding, painting, and other fabrication tasks.
US airlines American Airlines, Delta Air Lines, United Airlines, and Southwest Airlines have banned passengers from bringing humanoid robots that resemble humans or animals onto flights due to safety concerns over battery fire risks. Standard humanoid robots use lithium-ion batteries exceeding the FAA's maximum allowance of 160 Wh, posing catastrophic logistical and safety risks in a confined cabin. The bans explicitly close loopholes and clarify that humanoids do not qualify as emotional support devices.
Samsung is developing its own proprietary humanoid robot, integrating internal manufacturing data, custom hardware, and foundation AI models into a centralized effort. The company has established a dedicated Robotics eXperience (RX) Business Promotion Office reporting directly to CEO TM Roh and hired former Hyundai executive Lee Dong-geon to steer robotics strategy. Samsung's humanoid initiative is building a physical AI software stack combining its proprietary Gauss foundation model with new "world models" designed to simulate physical dynamics before execution.
A prototype has been developed to automatically check the ABI compatibility of ROS 2 core repository fixes, reducing the need for manual backporting efforts. The prototype uses two GitHub Actions: libabigail-action and ros2-abi-action, which compare shared libraries using libabigail and apply the REP-0009 policy for released distros and rolling. The tool can be integrated into the ROS 2 core repos and potentially shared with community packages, with feedback and concerns welcomed.
Industrial networks must adapt to increasing automation, with many devices producing and consuming data simultaneously. Network planning is now an operational issue, as a short interruption can halt a cell or delay diagnostics. To address this, teams should map traffic patterns, not just devices, and create a network that stays predictable as equipment and production requirements evolve.
Schaeffler has introduced a new high-speed forming process for strain wave gearboxes, reducing component costs by over 25% and material consumption by over 75%. This breakthrough will accelerate mass manufacturing in 2027 in Germany, followed by a global expansion to support humanoid robotics partners. The formed strain wave gearboxes complement Schaeffler's existing planetary gear actuators and secure the company's position as a Tier 1 supplier for humanoid robots.
Serve Robotics is expanding its autonomous delivery network through a partnership with Grubhub, adding a third major delivery platform to its network alongside Uber Eats and DoorDash. The Grubhub agreement is rolling out in Chicago, Los Angeles, and Alexandria, Virginia, with more than 100 participating restaurants in Chicago and nearly 200 in Los Angeles. Meanwhile, Diligent Robotics is launching its Moxi 2.0 mobile manipulation robot in hospitals across the US, with initial deployments in Chicago, Los Angeles, and Washington, DC.
Robots in dark or poorly lit environments rely on onboard lights, producing uneven illumination that degrades perception tasks. The authors extend Dark Gaussian Splatting (DarkGS) to create a more accurate and flexible relightable 3D reconstruction framework. This framework jointly optimizes lighting parameters within the Gaussian Splatting framework and incorporates a low-frequency illumination model based on spherical harmonics and an MLP-based Bidirectional Reflectance Distribution Function (BRDF) to model non-Lambertian reflectance.
MISTac, a high resolution vision-based tactile sensor, has been developed for minimally invasive surgery. The sensor has a replaceable sensor tip with a diameter of 8 mm and can resolve forces as little as 24.3 mN. It has been shown to be usable in minimally invasive surgery through an in vivo study, achieving an aggregate accuracy of ~84% in a tissue classification task. The sensor is now open-sourced for potential use in aiding surgeons during minimally invasive surgery.
A simulation model of an eel-inspired soft robot for anguilliform swimming is presented, using a Finite Element Method (FEM) model of an elastic rod to capture the soft materials of the robotic fish. The model is coupled with a hydrodynamic force model to simulate the behavior of the soft robot in water. The results show that slightly asymmetric designs are advantageous, offering comparable swimming velocities but greater maneuverability.
A real-time estimator of actuator control and health (REACH) algorithm for an eel-inspired soft robot is developed to accurately estimate actuator functionality in harsh underwater environments. The algorithm employs a soft robot model, sigma point filter, and formal statistical hypothesis test to capture nonlinearities and changes over time. REACH is evaluated using three sensor types (GPS, IMU, and Bend Sensor) and demonstrates that both bend sensor and IMU are adequate choices, with minimal differences in performance across three swimming gaits.
Recent 3D Gaussian Splatting (3DGS) has enabled efficient photorealistic view synthesis, but its offline reconstruction heuristics can become brittle under online SLAM's incremental map growth. Researchers proposed three geometry-aware methods for online mapping, including transmittance-preserving densification, camera-aware scale initialization, and error-guided densification, which improve rendering quality with negligible overhead.
Researchers have developed a task-aware, delay-extended discrete-time joint model for PMSM drives, enabling the direct synthesis of a discrete PI current controller with prescribed bandwidth and delay guarantees. The framework analyzes the minimum required sampling frequency, controller gains, and DC-link voltage needed to satisfy motor and joint performance, resulting in reduced sampling-frequency and DC-link-voltage requirements.
SkillComposer is an interactive natural-language robot programming system that learns reusable program abstractions. It uses a generate-test architecture with large language models (LLMs) to iteratively generate and revise robot programs before execution. Successful programs are stored and processed by an online library-learning algorithm that compresses recurring function sequences into reusable macro skills. Evaluations show that evaluator-guided generation and learned abstractions improve success rates and usability in natural-language robot programming.
GaussMemory, a task-driven 3D Gaussian scene memory, is proposed for long-horizon robotic manipulation, enabling a robot to actively learn how to remember objects and update its memory. It leverages 3D Gaussian Splatting as a persistent geometric substrate. GaussMemory outperforms existing memory systems in various benchmarks, including LIBERO and VLABench.