Robot ID codes and patrol robots

Robots moved deeper into regulated infrastructure today. China wants a permanent identity for every machine built on its soil, Ring wants a drone hovering in your hallway, and LiveKit wants your ROS graph reachable from anywhere. Below that, a jet maker inspects a fuselage with a humanoid and a carmaker teaches a robot to pick bolts out of a pile.

China drafts a lifetime ID code for every robot

The Ministry of Industry and Information Technology published draft measures on 6 November requiring a lifetime, tamper-proof identification code on every complete robot made in China, fixed before the robot leaves the factory (MIIT notice, in Chinese, Asia Tech News). Comments close on 5 December and the effective date reads “202X”.

The scope is broader than humanoids: industrial, service and special-purpose robots all fall under “one machine, one code”, stored digitally and marked so people and machines can read it. Manufacturers register each product, including key components and software, on a ministry platform and file the code within 20 working days. A humanoid standard is already precedent, 29 characters per robot, with more than 28,000 coded by May.

Industrial companies above the National Bureau of Statistics size threshold built 794,781 robots from January through September 2026, up 29.1 percent year on year. Europe is stricter about marking but later: the EU Machinery Regulation applies from 14 January 2027 and requires model or type, year of construction, and any batch or serial number.

The developer consequence is what has to become machine readable. A registry of components and software per unit assumes you can generate an as-built bill of materials, not assemble one by hand before an audit. The draft is silent on imports, fees and penalties, and no lookup API is announced, so watch for one and for how a software update to a coded robot gets recorded. That is governance landing on the device.

Ring puts a drone and a rover on patrol

Amazon announced Ring Robotics on 10 November with Ring Drone Cam, a flying indoor camera that docks out of sight, and Ring Rover Cam, a patrol robot for businesses (Amazon, TechCrunch).

Both run AI Pilot, combining device cameras, vision models, on-device sensors and cloud processing. You carry the drone through the house, or let the rover explore, and it turns what it sees into a floor plan, labels rooms and suggests patrol points. On patrol it plans a route and steers around furniture.

The drone has a 148 degree field of view, rotates 360 degrees, streams 2K and carries two LEDs for low light. The rover has a periscope mast saved to a height per stop, records up to 4K while moving, and adds a siren, light tower, two way audio and person detection when docked.

Indoor flight without GPS or a published lidar spec is the hard engineering, and Ring has not said which method it uses. The loop is public: semantic map, suggested waypoints, dock based recharge with the camera off, the map, plan, execute cycle Nav2 implements for AMRs.

Axon takes a fifth of Skydio

Axon invested $1.2 billion in Skydio, taking 20 percent ownership at a $5.2 billion valuation, and becomes exclusive distributor of Skydio drones to public safety agencies in the US and Canada (The Robot Report). The consideration mixes cash and Axon stock, split between primary capital and shares bought from existing holders, closing across the fourth quarter of 2026 and the first of 2027.

These are company figures. Skydio says it serves more than 1,500 public safety customers and expects about $1 billion in bookings this fiscal year. Axon estimates a $13 billion market for Drone as First Responder in the US and Canada, says drones arrive before officers on about 80 percent of the calls where they deploy, and puts response time cuts at up to 65 percent.

DFR is a scheduling problem wearing a flight controller: launch without a pilot, fly to a coordinate, hold a stream, return, recharge, while a control room decides which call gets the next aircraft.

Pilatus inspects a jet with a humanoid

Pilatus Aircraft is testing Hexagon Robotics’ AEON humanoid at its plant in Stans, Switzerland, where the robot performs quality checks inside the airframe of the PC-24 business jet (Assembly Magazine, Hexagon Robotics).

Roman Emmenegger, vice president of manufacturing, frames the goal as tighter inspections rather than headcount. AEON combines sensors, AI and precision measurement with motion control, and Hexagon positions it around spatial intelligence, manipulation, inspection and reality capture. The company has run comparable work with BMW and Schaeffler.

Inspection is a sensible first humanoid job because success is measurable: the scan either registers to the aircraft frame and matches nominal geometry or it does not. Reach and pose accuracy inside a fuselage are the hard parts, which makes this data capture as the product, not locomotion. Hexagon has published no accuracy figures.

A bolt detector that feeds a rule based gripper

Engineers at Dongfeng Motor’s research institute built a cell that picks loose bolts from a random pile at a 97 percent success rate and 8.73 seconds average cycle time (Assembly Magazine).

The vision problem is texture: bolts are small, elongated and reflective. The detector is a YOLOv8 variant with a depthwise grouped convolution in the C2f block, a lightweight channel attention module, a scale adaptive path aggregation neck, and a weighted distance IoU loss that tightens centre localization, which is what a grasp depends on.

The grasp is not learned. The bolts are known, the shank is the contact region, and a parallel jaw gripper clamping its midpoint symmetrically is a geometric rule. Across five bolt sizes, two end effectors and 1,000 pick and place operations, success ran from 98.5 percent on the largest bolt to 95 percent on the smallest, and held 96 to 98.7 percent with three to six bolts on the table. Grip slip on small or horizontal bolts was the dominant failure, camera depth error second.

When the object family is enumerable and the geometry is known, spend the effort on detection and validate the grasp analytically, the same trade we traced when end effectors bought generality and force bought throughput.

LiveKit puts the ROS 2 graph on WebRTC

LiveKit released Portal for ROS 2 on 10 November, a standalone ROS 2 node that joins a local graph to a LiveKit room so topics, service calls and video cross networks over WebRTC (LiveKit, announcement).

Topics travel as DataTracks with the message type discovered at runtime and a schema advertised alongside. Services route to a named participant through LiveKit RPC. sensor_msgs/msg/Image and CompressedImage become VideoTracks with configurable codec and bitrate.

Between two ROS graphs, messages stay in native CDR with no per message conversion, and Portal checks that both sides use the same message definition. To a browser or cloud service it publishes JSON plus a JSON Schema generated from the ROS type, which removes rosbridge and a hand written adapter. Topic rules live in YAML: regex patterns, direction, max_rate_hz, encoding, latched delivery and video options. DDS stays inside the robot network while LiveKit handles encryption, congestion control and selective subscriptions.

The sample config is version 0.0.1, so treat the surface as young. The risk is plain too: any topic exposed with direction: 'in' is a command channel, and those rules deserve the review you give the QoS profiles behind them.

Quick Bites

The RAI Institute published spot_basic, an MIT licensed Python layer over its spot_ros2 driver that wraps locomotion, manipulation, gripper control, transforms and camera streaming into one synchronous or asynchronous API (GitHub, ROS Discourse).

RoboACE evolves a curriculum harness against an execution harness while foundation models stay frozen, and reports 74.0 percent success on LIBERO-PRO, 15 points above its starting deployment (arXiv).

A Jazzy Jalisco sync is planned for Friday 13 November with 752 packages ready and no reported regressions on Ubuntu Noble; rosdistro pull requests are held until then (ROS Discourse).