There was a Cloud Robotics WG Meeting on August 24, 2026, from 4:00-5:00 PM UTC. The meeting continued editing the first draft of the Logging and Observability guide, which had been started in the previous session but not recorded due to lack of engaging video content. The next meeting link is available for sign-up or notification reminders through Google Group or calendar notifications.
Former Uber co-founder and CEO Travis Kalanick has raised $1.7 billion in equity investment for his new robotics startup ATOMS, with funding from top investors including a16z, Ben Horowitz, Bain Capital, and others. The funds will enable ATOMS to complete its "bits-to-atoms" story arc, focusing on building specialized robots for various industrial sectors such as food production, mining, and mobile robots. Lead investor a16z is joining the company's board alongside other prominent partners.
A free CLI tool that can convert CAD assemblies to URDF without requiring any CAD plugins is being considered for open-sourcing. The proposed tool would allow users to work directly with ROS from their terminal with just one command. Participants were asked if they would use such a tool, but also inquired about the biggest reason why they might prefer existing CAD add-ins instead.
A miniature robot, called "MIR" (Miniature Intraoral Robot), has been developed at the University of Basel to simplify dental treatment by precisely preparing teeth for a crown. The robot measures 43x26x28mm and can be fitted comfortably into an open mouth, reducing the need for multiple appointments with dentists. Researchers have tested the robot's precision, finding it to be less than 0.2 millimeters in positional error, which will be further reduced after sensors are integrated into the system. The next step is to integrate sensors and a camera to enable monitoring of its position and progress during treatment.
Holiday Robotics raised $105 million in Series A funding to build a wheeled humanoid robot called FRIDAY, designed to work safely alongside people in manufacturing environments. The robot features a wheel-based mobile base, dexterous hands with tactile sensing and force-aware control, and advanced perception systems. Holiday aims to increase reliability through its vision-language-skill approach, which combines hardware, simulation, learning, and deployment into a full-stack system.
AMD has unveiled the Kria module for real-time control and unified memory for robots, designed to deliver compute and AI workload processing for physical AI. The Ryzen AI Embedded X100 series and Kria AI Robotics platform offer deterministic real-time control, unified CPU-GPU-NPU memory, and an open software stack that can outperform NVIDIA's architecture on system-level robotics workloads.
HERMES, a heterogeneous edge-relational graph neural network with SSM-informed multi-head attention, was proposed for traffic conflict prediction at signalized intersections. Vehicles and pedestrians were represented as heterogeneous nodes, while vehicle-vehicle, vehicle-pedestrian, and pedestrian-pedestrian interactions were encoded as relation-specific edges. HERMES achieved an AUC-ROC of 0.9898, outperforming the strongest Transformer baseline and XGBoost, with a detection rate of 95.7% at a 5% false-alarm rate.
PhysCoRe presents a physics-corrected residual world model that combines a Material Point Method (MPM) simulator with two feed-forward neural networks. The design includes a material refinement module, which infers per-particle elasticity from visual observations, and a residual correction module, which learns the discrepancy between the simulator's internal dynamics and predicted corrections. This approach supports online material identification on novel objects and outperforms state-of-the-art baselines in prediction accuracy, providing a reliable distribution of confidence across an object's geometry. PhysCoRe achieves this by adapting its predictive uncertainty to steer further exploration toward regions where its estimate is least confident.
A low-cost, openly reproducible liquid-handling robot called RAINBOTTM was developed by converting a consumer-grade Cartesian 3D printer. The robot features a precision single-channel pipette actuated through the printer's X--Y--Z gantry, with plunger and tip-eject motions controlled by Python. A browser-based digital twin synchronises with the physical platform in real-time, allowing remote monitoring and intervention, and is coupled to the CEID framework for goal-directed inverse-design search. The complete hardware costs under US$1300 establish an accessible physical--virtual framework for self-driving laboratory automation.
Researchers have developed a learning-based framework for safe and scalable multi-drone cooperative payload transport using Discrete Graph Control Barrier Function Proximal Policy Optimization (DGPPO). The framework introduces a minimal 2D abstraction that preserves task-relevant drone-payload coupling while remaining computationally efficient. A single learned policy generalizes across varying team sizes and task scenarios, and demonstrates safe operation in dynamic environments with other drone teams acting as obstacles.
Researchers propose Capability-Aware Traversability (CAT), a framework embedding physical limits into spatial feature space. CAT grounds dense supervision masks in physical trajectories through an interactive annotation pipeline and modulates semantic terrain maps with robot-specific traversability vectors using Spatially-Adaptive Denormalization (SPADE) blocks. On human-annotated and trajectory-aligned datasets, CAT leads all ranking-based metrics, improving AUROC by 11.0% and AUPRC by 15.8%.
FELT, a learning-based framework, synthesizes per-finger pressure tactile images from RGB observations, reducing the need for tactile-equipped data collection. FELT uses a frozen visual encoder and a lightweight query decoder to predict tactile signals in a single feed-forward pass, capturing asymmetric contact patterns during interactions. At inference time, FELT only requires RGB data, allowing existing vision-only data to be augmented with tactile observations as generated tactile images or latent tactile features.
Researchers propose a decentralized UAV swarm system to protect against ground target attacks in GPS- and communication-denied environments. The swarm uses onboard sensors, Kalman filters, and relative measurements to track targets and coordinate as a unit. The approach is validated with real robots, demonstrating its effectiveness in detecting, encircling, and intercepting hostile UAVs.
Researchers Miroslav Krupa, Miroslav Cibula, and Kristína Malinovská have developed a self-supervised bio-inspired robotic trajectory planning framework with obstacle avoidance using forward and inverse models. The approach demonstrates feasibility but exploits the learning signal provided by the models, requiring additional training regimes and mitigation strategies to improve generalization.
A real-time RGB-D perception pipeline for autonomous impact hammers in mining simultaneously generates operationally feasible rock-breaking poses and a robot-free 3D representation of the workspace. The proposed approach combines image-based instance segmentation with geometric point cloud processing, operating on embedded hardware at approximately 10 Hz with a total latency of around 675 ms. This enables responsive closed-loop behavior when integrated with a control system. Experimental results in a representative scaled scenario demonstrate that the proposed system is suitable for real-time autonomous impact hammer operation.