LUMACTA NEWS / TOPICS
AI systems, models, agents, robotics and documented automation consequences.
ANALYSIS · AI & AUTOMATION
DeepMind says three-day track, intensity and wind-structure predictions matched the accuracy leading earlier systems reached at two days. Code and model weights are now public, but the benchmark is historical and official weather warnings remain authoritative.
REPORTED STORY · AI & AUTOMATION
JPL trained ERNEST's active-suspension controller in simulation and tested it on obstacle courses in the Mars Yard. A separate 26-kilometre desert campaign exercised longer-range autonomy with minimal intervention; integrating the two capabilities is the next research step. ERNEST is a research testbed; NASA/JPL's release does not announce a flight assignment.
REPORTED STORY · AI & AUTOMATION
Vision-language models generated waypoints for two Mars drives, and JPL engineers checked more than 500,000 telemetry variables for flight-software compatibility before uplink. The experiment changed ground planning; it did not hand unrestricted control to an onboard agent.