AM-IA brings together news, research papers and applications relating to artificial intelligence published around the world into a single feed — and makes them available in your European language at the click of a button.
Scroll to browse · Click to filter
Researchers propose a natural language planner that chooses skill, object and target, coupled with a push policy that achieves 93.3% success in simulation.
Driven by 1,616 demonstrations in calm and agitated water, vision-language-action models reach 93.3% success in the basin. On a wreck, an ACT policy succeeds 66.7% of a task.
According to The Information, the Max prototype weighs about 48 kg and must physically embody Meta's Agent Muse for domestic tasks. Meta says she doesn't plan on marketing it.
The robot Alice said the right words in the wrong order, or announced a visitor already present. A feedback offers TRACE, a review to identify these flaws.
Without any human demonstration or field adjustment, the robot captures the spreader in 90% of the phantom cases and maintains the gap when passing in 78%.
Gym-Anything for Robotics entrusts the creation of simulators to code agents, guided by US GDP. Average peak model score: 0.06.
The New York manufacturer of paillass robots, present with 11,000 machines in more than 80 countries, loops an E series led by Avego to accelerate "self-sustaining science".
Rather than a universal clamp, the University of Washington relies on interchangeable tactile adapters. Touch increases 15 to 25 points of success to policies learned by imitation.
By learning by strengthening, researchers teach a humanoid robot to avoid objects launched at high speed without losing balance or adopting unrealistic postures.
After more than 100,000 bins moved to Georgia, Digit will arrive in early 2027 at two GXO e-commerce sites. However, the Logistician estimates the return on investment to be two years.
This extra thumb, driven by a camera attached to the palm and driven by imitation, allows you to hold and manipulate with one hand. Hardware and software are published in open source.
Without tendons, all its motors housed in the palm, Torch-Hand aims to reconcile anatomical fidelity, gripping force and control simplicity for fine manipulation.
Rather than retrain a VLA model to avoid obstacles, a small policy learned by strengthening adjusts its movements on the fly: 52.8% successful tasks without collision, compared to 23.7% for the best alternative.
The MARL-Exo controller learned in front of a virtual body with various anatomys. Tested as it was in eight adults, it decreased their metabolic cost in all participants.
By rebuilding real scenes in simulation, Zero-HLM generates 4,800 physically validated demonstrations. The resulting policy succeeds 66.7% of tasks, compared with 46.7% with data collected on the real robot.
Driven on human demonstrations made without robot, with portable clamps that measure the effort, politics pilot both the gestures and the strength of each hand.
Boris Sofman's start-up, which transforms existing construction machinery into autonomous machines, more than doubled its valuation in eight months and signs eight new entrepreneurs.
The foundation model learns to walk, adjust its posture and manipulate with two hands from videos to the first person, and reaches 62.5% success in real-life conditions.
By reusing an already trained whole body controller, a team limits boxing learning to a small planning policy, trained in minutes.
Of the 39,800 seconds of actual driving with 13 models from three manufacturers, the track maintenance aid indicator shows only 47% of times when the perception of markings is degraded.