The ros2-apt-source package has been updated to version 1.3.0. This update primarily adds support for Fedora. Users with deploy scripts hardcoded to version 1.2.0 are advised to switch to 1.3.0. If the package is already installed, it will update automatically. No new features were added beyond this Fedora support.
Reports indicate a tension between US security concerns and its robotics industry's reliance on Chinese hardware, with US labs investigating Chinese LiDAR sensors. Engineers at Princeton have developed a motorless, origami-inspired robot that uses geometry for adaptable movement. Research on human-robot interaction suggests that expressive robots can trigger social responses to errors, such as releasing bonding hormones like oxytocin. The automotive sector is diversifying robotics adoption beyond cars, and events like the World Robot Conference showcase a wide range of industrial applications. Ultimately, managing expectations is crucial for successful robotics deployment, especially concerning human-robot interaction and real-world functionality.
Reporting obligations under the EU Cyber Resilience Act will begin on September 11, 2026, requiring manufacturers to report actively exploited vulnerabilities within 24 hours and full notifications within 72 hours. Open-source stewards will be covered starting December 11, 2027. The post raises a critical engineering question about distinguishing between exploited vulnerabilities and normal robot failures in the field. The author notes the difficulty of determining if an incident is an attack or a baseline failure without control runs. A need for established practices in the ROS world for baseline-rate logging is highlighted.
Phantom Bridge is recommended for interacting with ROS2 systems. It allows users to debug, visualize, and teleoperate any ROS2 robot. The post encourages readers to try this tool. An author plans to discuss the future and missing features at ROSCon 2026 in Toronto. The core functionality revolves around advanced control over ROS2 hardware.
The research proposes a novel criticality metric specifically tailored for vulnerable road users (VRUs) in autonomous driving. It introduces a scenario-independent framework to assess and predict criticality across all traffic participant classes, avoiding scenario-specific complexity. Evaluation on the DeepAccident dataset showed the VRU-centric metric improved pedestrian criticality classification by 50%. The prediction framework outperformed existing metrics by 275%, achieving an F1-score of 0.96. These results suggest a significant enhancement for safety evaluation in automated driving systems.
Pelican-Sim 1.0 is a general world model simulator designed for embodied intelligence, predicting future observations from visual context and robot actions. It features unified action representation, action-visual injection, sparse mixture-of-experts (MoE), and efficient rollout generation for improved control and speed. Training on large datasets resulted in significant gains in action controllability and video quality across various benchmarks. Downstream applications demonstrated strong policy success and performance improvements based on the simulator's predictions. The model exhibits qualitative generalization across shifts in trajectory, scene, object, embodiment, and viewpoint.
MoPA is a framework for coordinated mobile manipulation by aligning perception with mobility and manipulation while maintaining action coordination. It uses Dual Perceptual Streams with mutually masked query banks to extract separate perceptual representations from a shared context. Perception2Action Adaptation updates query banks and action streams via a Mixture-of-Transformers decoder, allowing information exchange between action streams. Coupled conditional flow matching learns a joint vector field for coordinated action generation. MoPA achieved state-of-the-art performance on the ManiSkill-HAB benchmark, demonstrating a mean full-task success rate of 76.3%.
A novel framework is introduced for pre-flight conflict detection in Advanced Air Mobility by combining uncertainty-conditioned trajectory prediction with polyhedra hazard representation. It uses a closed-form uncertainty estimation method integrating NURBS fitting with a Kalman Filter for state covariance propagation. Operator-conditioned weather hazards are represented as three-dimensional polyhedra volumes incorporating aircraft safety buffers. Conflict detection involves mesh intersection algorithms comparing the spatial uncertainty tube of the trajectory against these hazard polyhedra. This system allows for continuous strategic validation of flight plans as environmental conditions change.
RodForesight is a learning framework designed for slender and material agnostic rod insertion in precision manufacturing. It decomposes the task into a coarse approaching stage using visual servoing and a predictive insertion stage. During insertion, a diffusion policy generates actions while a world model predicts their effects on rod-hole alignment, allowing for pre-execution error evaluation. This method enables selection of the best action chunk based on predicted tilt and radial errors before execution. Experiments show that RodForesight significantly improves insertion success rates, increasing them from 88.9% to 96.7%.
The paper proposes a data-driven distributed control scheme using learning multi-objective Agent-Based Model Predictive Control (AMPC) for path-tracking. It addresses the challenges of handling multiple conflicting objectives in agent-based systems by decoupling the system using the Alternating Direction Method of Multipliers (ADMM). A learning-based method is introduced to initialize iterations and a data management method for real-time efficiency. The proposed scheme achieves the same control performance as the integrated method while reducing computational time by 43.5%. This learning approach further saves significant computational time, making it suitable for real-time implementation.
A physics-based closed-loop robotic bioprinting system is proposed for treating Volumetric Muscle Loss (VML). This system uses a 3D point cloud camera and vision algorithms to achieve near real-time parameter adaptation during printing. The system adjusts printing parameters to ensure a desired bead thickness is achieved based on quantitative evaluation. Experiments demonstrated that the system successfully converged to the target thickness with low tracking error and high repeatability. This framework offers a robust method for controlling bioprinting processes for tissue engineering applications.
The research proposes an automated, vision-based method to measure the thickness of bioprinted constructs (BPCs) with complex geometries. This method utilizes data from a structured light 3D camera, incorporating image-based segmentation and novel geometry-based thickness measurement algorithms on point cloud scans. The algorithms integrate segmentation masks with robotic forward kinematics and 3D point cloud data for precise thickness measurement. Evaluation showed high accuracy in simulation, with mean absolute errors ranging from 0.025 mm to 0.057 mm. The technique was successfully applied to a robotic bioprinting setup for real-world thickness assessment.