Today in Robotics — Tuesday, September 1, 2026
Today's spread runs from the money side of robotics to the messy physical world: a pre-seed video-generation model, a hard look at robot safety when attackers enter the picture, and drones learning to touch instead of look. We've also got humanoids working on their body language, defense-grade underwater autonomy, and DJI packing more than propellers for Berlin.
- 01 Visko launches Orbis live model and closes pre-seed funding round
Visko's Orbis claims 4K video at 24fps with hour-scale generation and no drift, pairing a technical debut with fresh pre-seed backing.
Read at source ↗ - 02 The Missing Layer in Robot Safety Assurance
A timely argument that safety assurance falls short unless it accounts for attacks that can corrupt how a robot sees, decides, and acts.
Read at source ↗ - 03 RoboGesture: Real-Time Semantic-aligned Co-Speech Gestures Generation for Humanoid Interaction
RoboGesture tackles real-time co-speech gestures for humanoids, aiming for movements that are both synchronized and semantically meaningful rather than merely decorative.
Read at source ↗ - 04 This missile-armed underwater drone could be quieter and harder to track than a submarine
Japan's plan to arm submarine drones—potentially quieter than crewed subs—has drawn Chinese criticism and signals where undersea autonomy is heading.
Read at source ↗ - 05 Watch: Drone trades cameras for touch to master the art of perching on branches
TU Delft's drone perches on branches using touch rather than cameras, a clever reminder that not every navigation problem needs vision.
Read at source ↗ - 06 US-made Vulcan drones enter production as Mobix acquisition nears
Vision Aerial pushes its American-built Vulcan family into production as the Mobix acquisition closes in, a bet on domestic manufacturing.
Read at source ↗ - 07 What to expect from DJI at IFA 2026
DJI arrives at IFA in Berlin with three newly launched products, signaling ambitions well beyond its drone roots.
Read at source ↗ - 08 RedLight-VLA: Models for traffic-rule grounding and behavioral emphasis in driving policies
RedLight-VLA grounds driving policies in traffic rules, targeting the rare signalized-intersection maneuvers that behavior-cloned models tend to fumble.
Read at source ↗
The RobotCafe digest — five minutes, the whole field.