SpaceX Caught Starship From Orbit
Ship 42 returned from orbit to the launch tower. The historic catch brings a fully reusable Starship closer, but recovery and reflight are different milestones.
Machines meet the real world. What works, what fails, and what comes next.
Ship 42 returned from orbit to the launch tower. The historic catch brings a fully reusable Starship closer, but recovery and reflight are different milestones.
In a simulated robot study, useful task advice improved retries. Matching that advice to the current mistake showed no detected advantage.
FLoRe teaches a humanoid how hard to press, not just where to move. Small controlled tests show why that extra command matters.
A portable teaching device captures a robot hand’s instructions as well as its movements. Small laboratory tests show why that distinction matters.
Three research models identify actions, describe what was attempted and judge the result. Their tests measure labels, not a robot’s newly learned skills.
A controller trained in simulation completed 173 of 237 physical insertions using joint motion alone. The trick works after contact, not across a whole factory.
A virtual lab caught a pipette movement that would not fit. Real dye experiments exposed what the simulation still missed.
GPT, Gemini and Claude controlled a real robot arm in eight tabletop tests. The results expose a gap between a good move and a finished job.
The LiteNWM study asks how much of the future a robot needs to predict. Its path evaluator improved goal-reaching in a small test, but added planning time.
A robot can hold fruit securely and still squeeze too hard. New research uses touch to adjust the grip, with a revealing trade-off in the results.
Softer footsteps, a wider detour, or both? A new robot study connects those choices, while revealing the time cost of being less disruptive.
IFR’s survey of 2025 sales puts transport and logistics at 47% of professional service-robot sales. That is a category lead, not a census of wheels versus legs.