Microbot Medical has partnered with Lovell Government Services. This partnership positions Lovell as a veteran-owned small business vendor for Microbot's federal programs. Through this arrangement, Microbot aims to increase veteran access to robotic surgery. The announcement was initially published on The Robot Report.
Developed a VS Code extension called "ROS2 Quick Runner" to streamline ROS 2 development within the editor. Users can right-click `.launch.py` files to automatically execute `ros2 launch`. Similarly, right-clicking Python or C++ nodes triggers `ros2 run`, handling workspace sourcing automatically. The extension also provides functions to source the workspace and build the entire ROS 2 workspace using `colcon build`. It intelligently detects the ROS 2 workspace by searching for directories containing package definitions.
Accurate field positioning requires not only a good GNSS receiver but also a robust correction network. The stability of the final position is affected by several factors, including station distance and signal quality. Furthermore, uptime and regional availability significantly impact positional accuracy. Utilizing a provider like RTKdata, which offers access to numerous stations, helps maintain strong Real-Time Kinematic (RTK) precision. Therefore, wide coverage of RTK stations is crucial for reliable surveying results.
Metrics possess a dual nature, capable of revealing valuable insights while simultaneously obscuring or corrupting reality. The author gained a deep appreciation for this complexity after over ten years of meticulously tracking his own life. This experience suggests a fundamental tension within the practice of measurement itself. Essentially, metrics are not definitive representations of truth.
RealSense introduced the D585 Pro, an AI-native depth camera designed specifically for robotic applications. This new model utilizes a proprietary Gen 5 system-on-chip for enhanced performance. It offers significantly improved depth quality, boasting over double the capability of previous RealSense generations. The announcement detailing this advanced sensor appeared originally on The Robot Report.
Kinova has introduced a new medical robotic arm named KIMA. This device was specifically designed for clinical applications, rather than being adapted from existing industrial technology. Details about the launch were initially published by The Robot Report. KIMA represents a dedicated development for use within medical settings.
Richtech Robotics is launching a livestream featuring their service robot named ADAM. This humanoid robot utilizes advanced Artificial Intelligence powered by NVIDIA technology. Users will have the opportunity to interact with ADAM during the broadcast. Information regarding this launch was initially published on The Robot Report.
MindOn Robotics unveiled a novel logistics demo showcasing one AI model managing several different robot types simultaneously. This system effectively unifies humanoids and dual-arm robots under centralized artificial intelligence control. A significant feature is its ability to operate without relying on pre-collected, specialized training data for each individual robot. The demonstration highlights an advancement in coordinating diverse robotic embodiments via a singular "mind."
Ros2 Robot is a PyQt/PySide6 based GUI designed to simplify ROS 2 development, functioning like "GitHub Desktop for ROS." In version 1.0.0, it allows users to manage packages by creating, rebuilding, and sourcing them with a single click. Key functionalities include running and managing nodes, real-time topic inspection, effortless ROS bag management, and building launch files using blocks. The tool also offers URDF visualization and plugin access while remaining educational about underlying CLI commands. It currently supports native Linux (Ubuntu 22.04/24.04) and Windows with WSL 2.
A student researcher is investigating the challenge of dependency blocking in robotics software development using ROS. They observe that engineers frequently halt progress when dependent modules, like localization or MoveIt, are unavailable or incomplete. To circumvent these blocks, developers currently write temporary scripts or use manual CLI commands to simulate missing components. The proposed solution is a ROS-native visual platform for mocking and managing calls to ROS 2 Actions and Services. Feedback is sought from experienced engineers regarding the pain points of dependency blocking and the potential value of such a development tool.
Robotics are transforming industries by enhancing efficiency, accuracy, and productivity across various sectors. Operations in manufacturing and logistics increasingly rely on automated systems for continuous function. This growing dependence necessitates reliable backup power sources to prevent operational downtime. Businesses seeking solutions for robotic infrastructure can find specialized information through Nite & Day Power. The company focuses specifically on uninterruptible power systems (UPS).
Developed a system called Physical Atari using a robot named Robotroller, which actsuate an Atari controller, and an Atari Devbox for rendering game frames and rewards. This platform allows studying reinforcement learning algorithms in a physical robotic embodiment alongside standard hardware like cameras. The system is designed to be robust through bearing-based movement and high-frequency servo monitoring, while remaining accessible using affordable, 3D-printable components costing under $1,000. Researchers used it for sustained experiments demonstrating that RL policies can learn directly on robots. Findings highlight the critical need for on-device adaptation due to performance degradation from minor distribution shifts between training and deployment.
WorkBenchMark is a new LEGO Duplo-based robotic assembly benchmark designed for the RoboCup Smart Manufacturing League. It addresses the challenge of combining low-level manipulation with task-level symbolic reasoning under physical constraints. The benchmark comprises 400 tasks distributed across four complexity tiers. Researchers are provided an open-vocabulary perception and an Assembly-by-Disassembly baseline solution. This planning-based pipeline demonstrated superior performance compared to a modern vision-language-action approach across all tested difficulty levels.
3D Scene Graphs (3DSGs) are powerful representations combining geometry with semantic relationships for spatial AI tasks in robotics and computer vision. Currently, the field is fragmented due to diverse formulations, construction pipelines, and evaluation protocols, hindering method comparison and deployment assessment. This survey aims to unify understanding by providing a critical review of 3DSGs under a common definition. It formalizes modeling choices, analyzing node/edge attributes and hierarchical structures. The work reviews how 3DSGs are built from sensory data and examines current downstream applications and evaluation strategies.
Introduces DiffusionVS, a novel diffusion-based framework for improving robust visual servoing in robotics. The method utilizes normalized image coordinates of observed tag corners to generate camera velocity via conditional denoising based on Diffusion Policy. To improve generalization beyond static datasets, it employs an online training paradigm that continuously expands training data through interactive experience collection. Simulations show near 100% success rates, and real-world experiments achieved a successful rate of 93%. Furthermore, integrating the diffusion mechanism demonstrably enhances performance across various existing visual servoing networks.
Current robot systems typically acquire reusable skills only through explicit, task-driven instructions. Playful Agentic Robot Learning introduces an approach where embodied coding agents utilize self-directed play for continual skill acquisition. Researchers developed Robotics Agent Teams (RATs) which propose exploratory tasks, execute policies, and diagnose failures during play. Successful executions are then distilled into a persistent code skill library for later use on new tasks. Experiments demonstrated that skills learned through this "play" stage significantly boost performance on held-out tasks compared to non-play baselines.
Novel applications of Resistive Force Theory (RFT) allow for approximating ground reaction forces when simulating robot movement in sand without detailed grain-by-grain modeling. Integrating 3D Granular RFT into the physics engine MuJoCo enables simulations for freely walking robots on such substrates. The developed tool successfully predicts key behaviors, such as foot sinkage and walking distance, with an accuracy of within 20\% compared to experimental results. Testing across various scenarios confirmed that crucial trends related to end effector shape, speed, and loading remain accurate. This open-source implementation shows promise for advancing the design of robots intended to traverse granular media.
Introduces an invariant extended Kalman filtering (InEKF) method for estimating humanoid robot states on moving ground using only internal proprioceptive sensors. This approach estimates the base position and velocity relative to the non-inertial ground frame, eliminating the need for external motion measurements. It utilizes kinematic constraints at the stance foot via foot-mounted IMUs to handle nonlinearities from ground motion. The resulting estimator has a right-invariant measurement model, beneficial when initial uncertainties are large. Experiments on swaying/pitching ground showed a 96% faster convergence and an 80% error reduction compared to prior InEKFs.
Large-scale autonomous agent systems challenge traditional formal methods. The Software Component Ensemble Language (SCEL) requires an improved operational semantics for global property reasoning. This report introduces a multi-layered model using category theory and sheaf theory to analyze SCEL. It posits that robot ensembles can be modeled as a sheaf over a topological space, linking components to points and knowledge to data. Consequently, processes like information sharing become equivalent to "gluing" local data, allowing system failures to be quantified as measurable topological obstructions.
SCAN-Planner is a novel, spatial collision-aware local planning framework designed for quadruped robots navigating complex environments. It addresses limitations of existing methods by using a yaw-aware twin-cylinder footprint to model the robot's elongated body for whole-body collision checks in 3D occupancy maps. The planner generates collision-free guidance using a projected A* search on an interpolated ground surface, which also manages vertical stability. For large-scale tasks, it incorporates a robot-centric sliding map that offers high-resolution local collision checking and recovery capabilities. Testing shows SCAN-Planner successfully produces safe, smooth, and efficient trajectories in dense clutter, unstructured 3D scenes, and stair traversal scenarios.
Developed pdSTL (probabilistic differentiable Signal Temporal Logic) to handle complex temporal specifications for robots in uncertain environments. This framework combines probabilistic semantics with differentiability over belief trajectories, addressing limitations of existing methods. It uses interval-valued probabilistic semantics to compute conservative satisfaction bounds within the STL syntax tree. The temporal robustness evaluation is formulated as a recurrent, LSTM-style unfolding, allowing efficient, differentiable monitoring. Validation on obstacle avoidance and quadcopter flight confirms pdSTL achieves optimization with formal probabilistic guarantees superior to deterministic approaches.