Teradyne–Sereact report physical-AI deployments in European logistics
The Robot Report says Teradyne Robotics and Sereact are partnering to expand physical AI in retail, logistics and manufacturing. It reports that hundreds of UR robots have already shipped as dual-arm systems to European customer sites, including CEVA Logistics sites handling Zalando returns in Germany and Poland. According to the report, Sereact's Cortex uses NVIDIA Isaac Sim and Cosmos for training, Isaac Lab for reinforcement-learning post-training, and Jetson Thor for inference at robot cells; DGX Spark supports site-level development and updates. Teradyne is also contributing UR and MiR SimReady OpenUSD assets to a Hugging Face library. These are reported deployments and architecture choices, not independently verified throughput or power benchmarks.
This edition covers finalized candidates from 2026-10-25 00:30 UTC to 2026-10-26 00:30 UTC (2026-10-25 08:30 to 2026-10-26 08:30 Asia/Shanghai). The frozen partial snapshot contains 7 finalized candidates; 5 selected or near-selected items are covered. Several items are developer posts or clipped summaries, so their claims and limitations remain attributed; this is not a comprehensive market survey.
Webots safety-gate test separates stopping from restarting
An Open Robotics Discourse author describes a solo Webots simulation with a TIAGo base: a UDP proposer requests 2.00 rad/s while a Checker alone writes actuator commands every 32 ms. A scripted 14.016-second run covered 438 cycles. Distress injected at 4.000 seconds caused a zero wheel command in the same cycle; restart required two separate 1.0-second dwell conditions and an explicit reset, accepted at 9.056 seconds. The author reports Python gate timing of p50 1.38 ms and maximum 4.50 ms, plus exact replay across 438 cycles. There was no live vision detection, human reset operator, industrial RTOS or safety certification; this is not evidence of certified real-world safety.
ROSCon China organizer reports community participation
An organizer recap says ROSCon China 2026 took place in Shanghai on October 16–18. The organizer reports more than 1,000 in-person attendees, over 300 represented organizations, 30 exhibitors and 50 keynotes or technical reports, alongside three courses, one workshop and one closed-door exchange. These are organizer-reported event counts, not measurements of ROS adoption or commercial deployments.
Robotheca developer launches a beta robotics catalogue
An Open Robotics Discourse developer describes Robotheca as a free beta catalogue and requests feedback. The developer reports 475 robot entries, more than 800 robotics software projects and more than 75 AI models. Described robot pages may include CAD/STL files, parts lists and prices where available, licences, build guides and ROS 2 or URDF tags. Licence labels reportedly come from project files, with ambiguous cases marked Unknown. These are developer claims about a beta directory, not independently audited coverage.
A source summary identifies an Arizona State University and University of Maryland paper, SEAL, on multitenant security-aware layer mapping for heterogeneous 3-D compute-in-memory accelerators. The visible abstract says stacked memory technologies for DNN inference may share power delivery and create multitenant security risks. The source says the researchers developed SEAL, but clipped fallback evidence does not expose the full algorithm or experimental measurements. No quantified mitigation benefit can be reported.