The US government has restricted imports of certain robots and connected power inverters produced in foreign countries due to concerns over national security. The Federal Communications Commission (FCC) determined that these products pose unacceptable risks to the national security of the United States, citing supply chain vulnerabilities and cybersecurity threats. Exemptions have been provided for advanced robots and power inverters that receive conditional approval from the US Department of War or Department of Homeland Security.
A company named 1X showcased robotic hands that are dexterous but also somewhat unappealing, with features like a "porn-pilled" translation tool and creepy glasses. In contrast, AI is having a positive impact on single chip workers in Korea who received giant bonuses due to its increasing value in the job market. The article highlights various AI-related advancements, including breakthroughs in language models and the development of faster chips needed for AI applications.
Shanghai Electric showcased its embodied intelligence solutions, including humanoids and industrial robots, at the 2026 World Artificial Intelligence Conference. The company's portfolio covers five key industrial scenarios and includes robots such as the "SUYUAN" bipedal humanoid robot with enhanced mobility, the "TUOYUAN" industrial wheeled humanoid robot, and the "Mermaid" bionic wheeled humanoid robot. Shanghai Electric also launched 51 AI models and agents, including an AI-native smart factory technology white paper, which proposes a real-time, closed-loop optimization architecture for production data.
Hoch Health Ostschweiz, a hospital in Switzerland, has expanded its modernization program with the integration of automated guided vehicles (AGVs) into its campus infrastructure. Schindler's elevators and digital solutions have been used to seamlessly integrate AGVs with elevator technologies, allowing for efficient transportation of laundry, food, medical equipment, and waste across large facilities. The hospital will be fully AGV-ready by 2027, with up to 50 AGVs operating throughout the campus, making it a leading adopter of automated hospital logistics in Europe.
Dexory has deployed SimScale's Engineering AI to accelerate engineering workflows for autonomous warehouse robots, aiming to reduce product development cycles while maintaining performance and reliability. The platform uses AI to identify root causes of structural component failures, explore design variations through simulation, and create a searchable knowledge base from previous projects. This deployment reflects a broader shift in manufacturing where AI is being embedded into engineering processes.
Humanoid robotics is transitioning to structured pilot deployments and early commercial adoption, driven by a 47% CAGR for humanoid unit sales over the next decade. Actuators are critical components responsible for motion, with electric actuators being the most popular among current robots, accounting for around 56% of a humanoid's weight. Soft actuators, using materials like electroactive polymers and shape memory alloys, are emerging as potential alternatives, but face challenges such as poor adaptability to unstructured environments. IDTechEx reports that material innovations will be key to developing next-generation humanoid hardware, with 10-year demand forecasts for key materials forecasted.
Shinkei Systems has introduced premium grade Alaska sockeye salmon under the Seremoni Grade brand. The company, behind seafood robotics and supply chain innovations, is expanding its offerings to one of the most widely enjoyed fish in the US. Shinkei's proprietary handling method produces exceptional flavor, texture, and shelf life. Seremoni Grade salmon will begin rolling out to select retail and restaurant partners this summer, marking a significant investment in Cook Inlet, Alaska.
Flytrex, an autonomous drone food delivery service, and Nash have partnered to build a multi-modal delivery platform that uses Flytrex's autonomous fleet and Nash's orchestration layer. The integration routes customer orders to the fastest, most reliable way to fulfill them, with drones handling the majority of orders and on-demand couriers carrying the rest. This partnership reflects a maturing view of how autonomous delivery scales, offering speed, cost, and reliability improvements for last-mile orders. Flytrex will remain autonomous-first, while Nash provides intelligence that decides the best mode for each order, regardless of weather or other conditions.
Wing and Walmart have launched a drone delivery service in Greater Orlando, marking Wing's first operations in Florida. The service enables customers to order thousands of everyday items from participating Walmart stores, with deliveries arriving in as little as 30 minutes. Initially, the service is available from two Walmart Supercenters in Apopka and Clermont, covering over 50,000 homes, businesses, and apartment buildings.
Renault Group has partnered with Exotec to automate its spare parts operation across France and Germany, reducing order processing time from 2 hours to just 15 minutes. The implementation of Exotec's Skypod robots enabled the company to consolidate over 80% of its French business into a single warehouse, significantly streamlining logistics. Renault Group has achieved customer service satisfaction rates exceeding 99.7%, timely deliveries and reduced costs through improved efficiency and productivity.
A systems framework for trustworthy embodied intelligence integrates four interdependent layers: the model layer, system layer, evidence layer, and deployment layer. The model layer generates task-competent action proposals with calibrated uncertainty, while the system layer realizes authorized actions dependably through integrated sensing, computation, control, and safeguards. The hierarchy of trustworthiness levels grades bounded deployment claims across task capability, safety, system assurance, operational governance, and supporting evidence, providing a basis for comparative evaluation, research prioritization, and future standardization.
Researchers have developed minimal-interface zero-shot agents, also known as agentic embodied control, that can rival industrial-scale policies in vision-and-language navigation. These agents sustain a long decision-making loop using only a monocular RGB camera and discrete actions, achieving 70.7% success in zero-shot navigation tasks. When a trained waypoint tool is exposed alongside primitives, the hybrid fable-5 agent reaches 76.7% success with half the environment steps and less than one quarter of the wall time of the maximum-effort primitive run.
A multi-objective framework, Marine Multi-Objective Compliance-Integrated Coevolution (MMOCIC), is introduced to balance team progress with norm violations in real-world deployments of learning-based coordination for collaborative robots. The framework blends coevolved behaviors with compliance behaviors and demonstrates high team performance while avoiding collisions on a swimmer rescue mission with up to 8 vehicles in hardware deployment, and 12 vehicles in simulation. The key contribution is the decoupling of learning from compliance since operational norms are prescribed rather than discovered. MMOCIC achieves this by processing sparse feedback signals to generate coordinated behaviors.
MoMo, a two-stage imitation-learning framework, uses a spatiotemporal action tokenizer and a behavior-cloning transformer to control motion modes in robot manipulation tasks. Across six real-robot tasks, varying MoMo's motion mode produces steady, dynamic, and intermediate behaviors that can be distinguished by human raters. These results demonstrate compositional generalization of motion modes across unseen task-mode combinations, showing that motion mode can be reused across tasks.
Researchers have developed a new framework for flexible needle pick up via thread manipulation to enable autonomous suturing, avoiding direct tool-tissue contact and enabling pickup even when the needle is occluded or inaccessible. The framework uses a suture thread as an assistive tool, reconstructing thread and tissue configuration, selecting safe grasp points, lifting threads stably, and following threads bimanually until securing needle grasping. This work evaluates the proposed framework on a da Vinci Research Kit under various real-world conditions, demonstrating robust performance even with challenging thread configurations or non-approachable needles.
A self-adaptive online learning method for tracking unknown target dynamics with no regret is proposed. The method simultaneously learns multiple predictors from scratch via self-supervised, one-shot, and computationally efficient learning, and adaptively selects the best one to match the observed target behavior. In the absence of error and switching, the method asymptotically matches the optimal non-causal control policy, while in the presence of errors and switching, it degrades gracefully with a regret proportional to the average learning error and switching times.
Researchers have developed a method to improve locomotion in cost-constrained quadrupedal hardware using reinforcement learning and a biologically inspired approach. By incorporating a forward model of the average actuator delay into a time-aware neural network, they achieve robust locomotion despite a 50ms transport delay. The proposed method also leads to the discovery of a central pattern generator (CPG) that mirrors the CPGs found in vertebrates, enabling the quadruped to adapt to latency perturbations. This approach may be a general strategy for cost-constrained locomotion.
RLMM-Flow combines expert flow-policy pretraining with latent-space reinforcement learning post-training to improve mobile manipulation framework quality. A flow policy captures a multimodal whole-body motion prior from expert demonstrations, and a latent steering network steers its initial noise toward higher-value action chunks. Joint training of the latent critic and actor improves task success, collision avoidance, and trajectory quality over imitation-only flow policies and existing reinforcement learning post-training baselines.
Researchers developed a Concept Expert module to build executable Analytic Concepts representing objects as explicit, programmatic blueprints. This mechanism bridges the gap between 2D Vision-Language-Action (VLA) models and the rich object structural information in the 3D physical world. The Concept Expert leverages 3D information from Vision Foundation Models (VFMs) to estimate initial kinematic and structural parameters before VLA inference, ensuring persistent accuracy throughout manipulation.
Researchers developed Speech2Grasp, a framework to transfer text-conditioned grasp detection to speech in humanoid robots, improving upon existing approaches. Using ALBEF as a case study, they found that adapting the model to speech using a lightweight MLP-based projector preserves semantic discrimination and robustness. The real-world experiments showed that Speech2Grasp outperforms traditional cascaded ASR-based pipelines while reducing inference latency.