A new system called SceneSmith, developed at MIT CSAIL and Toyota Research Institute, uses three agents to create realistic 3D scenes that can be used to train robots. The system's virtual environments are generated using a multi-modal vision-language model (VLM) called GPT-5.2, which is trained on internet text and images. These virtual playgrounds allow robots to practice skills and try out different ways of doing tasks before they're powered on, reducing the need for real-world testing. The scenes created by SceneSmith are more realistic and detailed than prior systems, with a high success rate in evaluating robot actions and improving roboticists' approaches.
The ROS 2 Client Library Performance Monitoring project aims to connect benchmark execution, result processing, and visualization. The CLI workflow runs a reduced rclcpp benchmark matrix, converts raw outputs into JSONL format, and supports multiple communication modes and layouts. A public dashboard provides interactive analysis with features such as latency, throughput, CPU usage, memory usage, and message reliability comparison views. The project is still in progress, focusing on improving repeated run summaries and handling incomplete upstream benchmark data.
Brain Corp reports a 68% year-over-year growth in global deployments of BrainOS-powered robots, with over 5.3 million autonomous operating hours logged in the first half of 2026, covering more than 3.7 billion square feet. The company has surpassed 50,000 robots deployed worldwide and is marking an inflection point for enterprise robotics as organizations increasingly adopt proven platforms for deployment and scalability. Brain Corp's global fleet footprint points to a growing role of autonomous robots in various industries, including retail, logistics, airports, and commercial cleaning.
Regal Rexnord and Fanuc have formed a strategic collaboration to deliver a fully integrated collaborative robotics solution, combining Thomson linear motion technology with Boston Gear gearheads, Huco couplings, Kollmorgen motors, and software. The solution enables Fanuc CRX users to deploy the Thomson Movotrak Cobot Transfer Unit (CTU) as an external axis solution, extending cobot reach, collision detection capabilities, and productivity.
Tate, a 120-year-old data center infrastructure company, has deployed a fleet of 58 Hirebotics cobot welding systems across its manufacturing facilities in Arkansas, Virginia, and Kentucky. The systems have achieved a 12x increase in per-welder throughput on critical structural assemblies, while expanding the team's certified welders. Hirebotics' Beacon Pro control platform enables Tate to program, run, and monitor the cobots from any smartphone or tablet, improving production efficiency and shop floor conditions.
Icarus Robotics is building a "robotic labor force for space" by combining embodied artificial intelligence with human-in-the-loop control to automate routine tasks in orbit. The company aims to free astronauts from repetitive jobs such as cargo handling, equipment inspection, and experiment setup, allowing them to focus on scientific research and exploration. Icarus' technology will eventually support commercial space stations, orbital manufacturing facilities, and future missions to the Moon and Mars.
Matthäi Schlüsselfertigbau, a subsidiary of the German construction giant Matthäi, has established its own 3D printing division after successfully completing a new two-story headquarters printed with a COBOD BOD2 printer. The building was constructed in about 105 hours and provides 364 m² (3,920 SF) of usable space. The decision follows the company's growing interest in innovation and progress in construction, as it combines entrepreneurial growth with technological advancements.
PIA Automation is developing an E-Drive assembly line at the BMW Group plant in Steyr, Austria, featuring 46 industrial robots and digital twin technology across 5,600 square meters. The system assembles transmissions for electric vehicles before joining them with electric motors to create complete E-Drive units. PIA Automation applied customized automation solutions, digital twin technology, and Virtual Commissioning throughout the project's engineering and implementation.
Analog Devices has renewed its partnership with MassRobotics, a Boston-based robotics accelerator, as a strategic sponsor. Analog Garage, ADI's internal corporate venture and incubator group, will provide seed funding, engineering resources, and technologies to support startups in developing production-ready realities. MassRobotics startups gain access to ADI's advanced sensor, signal processing, and power management technologies, with the goal of accelerating hardware development for complex robotics systems. Analog Devices will participate in various events and conferences, including the RoboBusiness 2026 Robotics Startup Bootcamp session and IROS 2026 Startup PitchFest.
Multiway Robotics has completed the deployment of an intelligent warehouse automation system for a Malaysian manufacturer, combining autonomous forklifts with warehouse management software. The system supports over 5,000 storage locations and has increased storage density by more than 30 percent while providing end-to-end automation for warehouse operations. Three autonomous forklift models are deployed throughout the facility, including the MW-K16 VNA forklift that operates in narrow aisles and can lift pallets to heights of up to nine meters. The system provides real-time visibility of inventory and coordinates fleet movements, optimizing routes and managing automatic charging.
VLAff, a vision-language model-based unified foundation model, learns cross-modal correlations across actionable affordances, including visual affordance heatmaps, grasp poses, and trajectories. This allows VLAff to generate directly executable actions from visual observations and instructions. VLAff achieves state-of-the-art performance on visual affordance prediction and demonstrates effective application in real robot applications, such as zero-shot manipulation and affordance-guided robot learning. The model utilizes 3D scene information for conversion into executable actions.
Researchers Duc M. Nguyen and colleagues introduce a novel safety formulation to evaluate the probability and severity of impactful interactions between robots and surrounding entities during failures, enabling informed planning decisions that balance safety with task efficiency. They also present FailBench, a MuJoCo-based simulation framework for studying robot-environment interactions under diverse failure modes. The approach provides a foundation for developing safer and more robust motion plans and learned policies in real-world household environments.
World-to-Wrist: Task-Conditioned Future Wrist Modeling for Fine-Grained Robot Manipulation presents a VLA model that contextualizes wrist local interactions under global task context. The model forecasts future wrist latents, transforming them into future-aware context for action prediction. Experiments demonstrate improved fine-grained and contact-sensitive manipulation across single-arm and bimanual settings, with action-generation rates above 80 Hz.
A concept hardware design has been presented to enable mechanically reconfigurable sensing in continuum robots, allowing uncertainty-shaping in state estimation. A sensor can be longitudinally translated within the robot, demonstrating a reduction of full-body shape estimation errors when sliding the sensor back and forth over time. This approach enables achieving sufficient confidence at task-relevant locations along the robot, prioritizing accuracy where it is most needed.
ARGUS, an observation pre-processing pipeline, uses large-scale 3D vision models to align image observations from arbitrary camera viewpoints into a canonical viewpoint before passing it to downstream visuomotor policies. This approach outperforms prior methods in efficiency comparisons and consistently improves success rates when learning from view-diverse data, converging up to 4-6x faster than previous methods by leveraging a simplified observation space.
PathCover, a framework driven by Randomized Iterative Space Partitioning (RISP), generates obstacle-free regions for autonomous robot navigation. It constructs convex polytopes directly from raw point cloud data in expected linear time and achieves an order-of-magnitude speedup over state-of-the-art methods while maintaining comparable corridor volumes.
KILVO (Kinematic-Inertial-LiDAR-Visual Odometry with Robust Multimodal Adaptation for Humanoid Robots) is a odometry system tailored to humanoid robots, utilizing joint encoders, IMU, LiDAR, and camera sensors within an asynchronous-sequential hybrid error-state iterated Kalman filter (ESIKF). The framework achieves highly competitive accuracy, efficiency, and output rates with strong robustness against sensor degradation and failures.