Einride, an autonomous freight developer based in Stockholm, is publicly listing through a SPAC merger. The company operates a fleet of electric, heavy-duty trucks. Its services provide logistics support to major Global Fortune 500 companies. Operations are currently active across both the United States and Europe. This news was initially reported by The Robot Report.
World-Action Models (WAMs) represent an advancement in robot policy design. These models begin with a pre-trained Vision-Language Model (VLM) backbone for initial capabilities. They are then fine-tuned to translate high-level linguistic instructions into physical actions. Essentially, WAMs bridge the gap between understanding natural language and executing complex physical tasks. This architecture allows robots to move beyond simple recognition toward goal-directed action planning.
Built Robotics and Penn xLAB are collaborating to advance artificial intelligence in the construction industry. This partnership aims to collect more extensive data specifically related to construction sites. The gathered data will be used to improve existing AI models. A key focus of this development is enhancing safety protocols within construction environments. Through this effort, they plan to create physical, functional AI applications for the sector.
PSYONIC is collaborating with ABB Robotics, integrating the company’s Ability Hand with ABB's GoFa collaborative robot arm. This partnership aims to translate data from prosthetic users into advanced robotic grasping capabilities. By combining these technologies, the effort seeks to imbue robots with a more human-like dexterity and touch. The news regarding this collaboration was originally published on The Robot Report.
Local ROS Meetups were successfully held in Heilbronn (March) and Karlsruhe (May), demonstrating a robust grassroots interest in the ROS 2 ecosystem. The Heilbronn event featured diverse technical presentations on topics like real-time control, motion planning, and LLM-controlled robots. The Karlsruhe meetup covered broad applications from medical robotics to integrating ROS with industrial CNC machines. Both gatherings underscored the continued importance of face-to-face networking for advancing open-source robotics development. These successful events highlight a growing regional appetite for collaborative knowledge exchange within the ROS community.
A user with experience in C, Python, ROS2, and control theory is seeking a stable repository or project within the ROS2 ecosystem to contribute to. The individual has provided a GitHub link for potential reviewers to assess their suitability for contributions. Recommendations for suitable starting points are requested by the user. The goal is to become actively involved in developing or supporting existing ROS2 projects.
AutoFlight's V2000CG CarryAll has received a Validated Type Certificate (VTC) from Indonesia. This makes it the first eVTOL globally to obtain overseas type certificate validation. The certification confirms the aircraft’s design meets airworthiness standards for two major bodies. Specifically, compliance was verified with both China's CAAC and Indonesia’s DGCA.
The global mining industry is undergoing a major transformation powered by artificial intelligence, robotics, and full automation. Intelligent excavators and autonomous mining trucks are now standard equipment in modern mines. These technological advancements significantly boost both operational safety and overall production efficiency. Consequently, the standards and demands for professional mining machinery parts have also risen considerably.
A map-free framework integrating reactive rolling-horizon planning with nonlinear Model Predictive Control (MPC) addresses autonomous navigation in unknown settings. The system uses LiDAR data to create Gaussian occupancy maps, generating collision-free trajectories via A* search tracked by an MPC formulation. Robustness against parameter sensitivity is enhanced using an offline Bayesian optimization scheme based on Tree-structured Parzen Estimators (TPE). A Gaussian Process surrogate aids in analyzing the resulting parameter landscape. Evaluation on a Unitree Go2 quadruped demonstrated that simulation-tuned parameters maintained high performance on hardware, achieving up to 90.0% navigation success.
Autonomous Uncrewed Aerial Vehicle (UAV) safe landings in unstructured areas require precise semantic understanding despite limited annotated aerial data. The proposed solution introduces a perception and data generation pipeline tackling the sim-to-real gap. This system uses a procedural synthetic engine to create photorealistic, automatically annotated urban environments through domain randomization. A OneFormer Transformer is fine-tuned on this generated data for global context resolution in segmentation. Finally, a deterministic module employs an Euclidean Distance Transform to identify safe landing zones with guaranteed obstacle clearance, proving effectiveness via UAVid benchmarking and real-world validation.
Accurate position estimation for autonomous lunar landings requires robust Terrain Relative Navigation (TRN). A proposed algorithm integrates a deep-learning crater detector with an Extended Kalman Filter (EKF). The detector analyzes monocular images from orbit, matching observed craters against a global database using Hungarian assignment and outlier removal. These identified measurements refine the EKF, estimating spacecraft pose in the LCLF frame while constraining radial drift with altitude aiding. Simulations show the TRN can recover navigation accuracy to hundreds of meters even after an initial positional error of up to 5 km.
To allow assistive robots to adapt to individual user tastes, the paper proposes using language to reason about preferred behaviors rather than directly generating them. The method learns a structured latent representation that organizes these user preferences based on trajectory differences. When given a preference prompt, a foundation model interprets this latent space to select a specific value corresponding to the desired behavior. Experiments demonstrate that selecting robot behaviors from this structured latent space enables more precise adaptation to user tastes. This approach achieves better results while requiring substantially fewer preference labels than direct language-conditioned policies.
VANDERER is an exploration framework designed for mobile agents to navigate unseen environments without relying on accurate occupancy maps, which are difficult to generate with limited sensors like monocular cameras. It guides pre-trained diffusion policies by incorporating a Visual Curiosity Module (VCM). This VCM utilizes a navigation world model to predict the outcomes of potential actions and quantifies them using a curiosity cost. The resulting cost actively steers the diffusion process toward generating actions that maximize exploration within the environment. Empirical testing across diverse simulated settings showed VANDERER significantly surpassed existing methods, exploring an average of 13.4% more area than NoMaD.
Investigated physiological responses during contact-rich physical human-robot interaction across 18 environmental combinations varying temperature, noise, and light. Multimodal data—including EDA, sEMG, eye-tracking, and subjective ratings—were collected alongside task performance metrics. Results indicated that while objective performance remained constant regardless of environmental stress, autonomic workload (SCL) increased with elevated temperatures. This suggests operators compensate for thermal discomfort by increasing physiological effort to maintain consistent execution. These findings recommend developing control systems that incorporate real-time physiological data to actively reduce operator workload in complex environments.
ViTaL is a framework that steers pre-trained generative robot policies during deployment by using both visual and tactile information for action verification. It models multimodal guidance as a bi-level optimization problem: high-level visual sampling selects the overall behavior, while low-level tactile diffusion editing refines the action sequence to meet local contact requirements. To support outcome-based steering, ViTaL learns a visuo-tactile latent world model and utilizes semantically aligned verifiers, including a text-conditioned tactile reward in latent space. Testing across three real-world, contact-rich tasks showed that ViTaL increased overall success by 51% compared to the base policy. The framework notably outperformed unimodal steering approaches by at least 33%.
LV-Calib is a new framework for estimating LiDAR-camera extrinsic parameters and performing LiDAR boundary-response calibration using a printable planar target. The target simultaneously provides indexed visual measurements via fiducials and structural feature points from circular reflectivity boundaries detectable by LiDAR. Instead of assuming ideal contours, the method refines accurate 3D LiDAR features by handling background point cropping and the broadened transition band at discontinuities. This leads to a weighted reprojection-consistent extrinsic optimization using image observations and refined LiDAR feature alignment. Finally, LV-Calib calibrates the LiDAR boundary response by estimating pitch-yaw-range residual statistics from overlapping boundary samples, achieving high accuracy in experiments.